Imaging Pearls ❯ August 2025
|
-- OR -- |
|
3D and Workflow
- Traditionally, medicine has relied on objective data to drive progress and guide adaptation across clinical care, research, health administration, and policy. However, if we hope to continuously accelerate how we adapt health care over time, true progress demands more than data—it requires a strategic embrace of risk and resilience. Those terms may seem at odds with the field’s emphasis on safety and certainty, given the high stakes of human life. Yet, there are ways to facilitate calculated risk and foster true scientific resilience in the effort to innovate, improve, and ultimately better preserve human life.
Data, Risk, Adaptation, and Resilience in Modern Health Care.
Giovanis T, Fishman EK, Chu LC, Rowe SP, Crawford CK.
J Am Coll Radiol. 2025 Jun 19:S1546-1440(25)00338-2. - Challenging the systems that uphold health and life may seem risky, but it is precisely that scrutiny that drives meaningful improvement. We must constantly re-evaluate the efficacy of our current policies and how they impact health care, not only on a national level, but right down to the level of the hospital and patient. This is not to say that current policies are inherently flawed, but rather that dynamic environments, such as health care, require dynamic oversight and continuous adaptation. Given the scale and complexity of today’s health care systems, fine-tuning their operations is a formidable task.
Data, Risk, Adaptation, and Resilience in Modern Health Care.
Giovanis T, Fishman EK, Chu LC, Rowe SP, Crawford CK.
J Am Coll Radiol. 2025 Jun 19:S1546-1440(25)00338-2. - The National Institutes of Health and National Cancer Institute are just two examples of the federal institutions that provide funding for promising initiatives all across the country; however, approval may take 7 to 10 months or more. Although these lengthy processes ensure due diligence and minimize risk, they can also delay timely responses to emerging scientific needs. Further, the general aversion to risk at such grant funding agencies can lead to a concentration of funding in established programs or established investigators and freeze out potentially higher-risk and higher-reward proposals from new and innovative researchers.
Data, Risk, Adaptation, and Resilience in Modern Health Care.
Giovanis T, Fishman EK, Chu LC, Rowe SP, Crawford CK.
J Am Coll Radiol. 2025 Jun 19:S1546-1440(25)00338-2. - In contrast, private sources of funding can bypass the lengthy bureaucratic procedures, enabling them to take calculated financial risks on promising research with far greater speed. Although the government takes 10 months to process grants, privately funded operations with clear initiatives can process grants in just days to a week. Private organizations’ agility is due to three unique advantages that allow them to operate swiftly and decisively with the freedom to take more risks on high-potential projects that might not survive the traditional selection processes.
Data, Risk, Adaptation, and Resilience in Modern Health Care.
Giovanis T, Fishman EK, Chu LC, Rowe SP, Crawford CK.
J Am Coll Radiol. 2025 Jun 19:S1546-1440(25)00338-2. - First, private organizations can streamline the approval process by avoiding the multiple layers of review typically involved in government review panels. In addition, organizations can specialize in certain areas, narrowing their scope and focus to further accelerate the approval speed and risks they are willing to take. Although government grant funding agencies may have multiple strong personalities in a given study section— each of whom can work to exclude certain promising grants from consideration for funding—private organizations may rely on a very select cadre of domain experts or passionate advocates to greenlight innovative grants.
Data, Risk, Adaptation, and Resilience in Modern Health Care.
Giovanis T, Fishman EK, Chu LC, Rowe SP, Crawford CK.
J Am Coll Radiol. 2025 Jun 19:S1546-1440(25)00338-2. - “Second, private organizations possess the unique advantage of being able to invest not only in projects but also in the individuals behind them. It is essential to acknowledge that the effectiveness of research is fundamentally tied to the capabilities of the team conducting it. Without access to top-tier talent, even the most important initiatives risk falling short of their goals.Young and innovative investigators with game-changing ideas may find a more facile path via funding from private organizations.”
Data, Risk, Adaptation, and Resilience in Modern Health Care.
Giovanis T, Fishman EK, Chu LC, Rowe SP, Crawford CK.
J Am Coll Radiol. 2025 Jun 19:S1546-1440(25)00338-2. - Last, streamlining the funding process and investing in both people and projects can foster closer, more collaborative relationships between funders and researchers. Although some organizations may prefer a handsoff approach, the increased engagement enabled by an accelerated process often leads to a deeper understanding of the research, stronger partnerships, and enhanced accountability. Although independent funding sources cannot replace the scale of public funding provided by the federal government, they can serve as a critical supplement, supporting targeted areas of research and enabling the kind of bold, innovative strategies essential for accelerating progress in health care today. And, in the current funding environment, private organizations, including both philanthropic groups and pharmaceutical industry sponsors, may come to fill the gap left by decreasing federal budgets.
Data, Risk, Adaptation, and Resilience in Modern Health Care.
Giovanis T, Fishman EK, Chu LC, Rowe SP, Crawford CK.
J Am Coll Radiol. 2025 Jun 19:S1546-1440(25)00338-2. - In an era in which health care must continuously evolve to meet the growing and shifting needs of society, embracing adaptability through calculated risk and fostering a resilient complementary system are no longer optional—they are essential. Accelerated innovation demands the courage to question existing policy frameworks and pursue research through more agile and responsive models. By reevaluating health policy with a critical yet constructive eye, and by supporting flexible, risk-tolerant research through private capabilities, we can create a more dynamic, future-proof health care system. The health care system must be prepared not only to respond to current challenges but to proactively reshape itself for the future of medicine. In blending the reliability of established science with the boldness of risk, the groundwork is laid for a more resilient, more responsive, and ultimately more effective approach to improving human health.
Data, Risk, Adaptation, and Resilience in Modern Health Care.
Giovanis T, Fishman EK, Chu LC, Rowe SP, Crawford CK.
J Am Coll Radiol. 2025 Jun 19:S1546-1440(25)00338-2. - One of the things we should excel at is the ability to explain how our clinical activities help us define research that is truly worthwhile and truly of patient benefit. Competing with “pure” researchers while simply presenting similar Material is never going to be the path to success. The novel thoughts that are presented in this article suggest we need to think differently. Whether thinking differently may lead to success, it will still take hard work and drive. Looking for unconventional sources of funding may be the answer. Whether it is foundations that support a specific disease (ie, The Lustgarten Foundation and Pancreatic Cancer) or diseases (Mark Foundation and cancer research) or organizations that focus more on an individual than simply the project will also be important for radiology clinical researchers.
Data, Risk, Adaptation, and Resilience in Modern Health Care.
Giovanis T, Fishman EK, Chu LC, Rowe SP, Crawford CK.
J Am Coll Radiol. 2025 Jun 19:S1546-1440(25)00338-2. 
- Although job displacement due to automation is not new, AI is accelerating this trend in ways that were not possible before. Industries like primary metal production have seen significant job losses due to automation; between 1990 and 2018, jobs in primary metal-producing industries like iron and steel manufacturing, production and processing of aluminum, and foundries contracted by 328,000 [2]. When compared with the production index for the same period and industry, the average production level remained unchanged relative to the US Federal Reserve Board benchmark [3]. Thus, automation can affect jobs without negatively impacting overall output— a pattern we are now seeing in many other industries due to AI’s rise.
From Automation to Innovation: How Artificial Intelligence Is Reshaping Global Industries.
Smith J, Fishman EK, Chu LC, Rowe SP, Crawford CK.
J Am Coll Radiol. 2025 Jun 19:S1546-1440(25)00340-0. - “As we move toward 2030, McKinsey & Company predict that 375 million people, or 14% of the current global workforce, may need to switch jobs . In addition, PwC’s Global Artificial Intelligence Study predicts the potential contribution of AI to the global economy will total roughly $15.7 trillion, or 26% in global GDP boost by 2030 [5]. Both estimates and the historical precedence underscore that job impact and displacement is real—something that for which job markets, both blue and white collar, should be prepared.”
From Automation to Innovation: How Artificial Intelligence Is Reshaping Global Industries.
Smith J, Fishman EK, Chu LC, Rowe SP, Crawford CK.
J Am Coll Radiol. 2025 Jun 19:S1546-1440(25)00340-0. - While analyzing the degree of simultaneous disruption and opportunity that AI will have on labor markets, it is reasonable to assume that jobs requiring more creative or interpretative skills will be less affected by the robust computing and repetitive skillsets that AI algorithms possess. In addition, as expected, we now see career expansion in technological or digital complementary professions, such as data analysts, machine learning analysts, software engineers, and cybersecurity experts. As AI evolution progresses, individuals may want to closely observe cues and signals from outside fields to gauge the impact that AI might have on their field.
From Automation to Innovation: How Artificial Intelligence Is Reshaping Global Industries.
Smith J, Fishman EK, Chu LC, Rowe SP, Crawford CK.
J Am Coll Radiol. 2025 Jun 19:S1546-1440(25)00340-0. - With the more recent emergence and commercialization of generative AI, we are starting to see how creative markets can leverage and apply AI to their work. Image-generating programs andmusical composition AI is starting to gain traction in concept generation or other works that have traditionally been under the direction of creative directors or similar professionals. Generative technology is continuously improving over time, and while human expertise and prompts are still required for function, the dissolving technical prerequisites seem as if the field is going to open up to a wider range of creative professionals.
From Automation to Innovation: How Artificial Intelligence Is Reshaping Global Industries.
Smith J, Fishman EK, Chu LC, Rowe SP, Crawford CK.
J Am Coll Radiol. 2025 Jun 19:S1546-1440(25)00340-0. - Although finance is performing actions based on algorithmic cues and the arts are generating from prompts, journalism is ramping up production by combining the applications of AI from Both fields. The Associated Press is now producing 4,400 earnings stories per quarter in the same amount of time it formerly took to do 300 stories. With that integrated strategy, coined “The Robot Reporter” by the Associated Press, input and data are compiled, analyzed, and then transformed into a story ready to be published. That follows the same process a human journalist would use to produce a similar original piece but with the added productive capability of AI systems and large language models.
From Automation to Innovation: How Artificial Intelligence Is Reshaping Global Industries.
Smith J, Fishman EK, Chu LC, Rowe SP, Crawford CK.
J Am Coll Radiol. 2025 Jun 19:S1546-1440(25)00340-0. - The evolution of AI is fundamentally reshaping fields across the globe, from finance and legal to creative and marketing sectors. Although AI offers immense opportunities, such as enhanced productivity and the creation of new roles, it also poses challenges, particularly regarding job displacement. Ultimately, embracing AI’s potential will require industries to strike a balance between automation and the irreplaceable value of human creativity and judgment. As AI continues to develop, its influence will only grow, adjusting how we work, create, and interact with technology every day.
From Automation to Innovation: How Artificial Intelligence Is Reshaping Global Industries.
Smith J, Fishman EK, Chu LC, Rowe SP, Crawford CK.
J Am Coll Radiol. 2025 Jun 19:S1546-1440(25)00340-0. - “To date, AI has not impacted medicine in the same way that automation impacted industries such as primary metal production in the last century. Instead, AI has helped “level up” medical professionals, including radiologists. That is particularly evident in mundane tasks that require significant focus, such as radiologists painstakingly searching for pulmonary nodules or subtle pulmonary emboli. As with many other fields, the current capabilities of AI are best suited to tackle such large-scale, repetitive tasks that lack creative or dynamic thinking. However, as generative AI (eg, large-language models, diffusion models) continues to evolve, many other tasks start to fall within the domain of AI, including complex and “creative” tasks such as image interpretation and dictation generation.’
From Automation to Innovation: How Artificial Intelligence Is Reshaping Global Industries.
Smith J, Fishman EK, Chu LC, Rowe SP, Crawford CK.
J Am Coll Radiol. 2025 Jun 19:S1546-1440(25)00340-0. - If clinical documentation can be handled effectively by AI—something demonstrated already by AI assistants, which can compile realtime notes from information obtained during clinic visits—then time and resources can be better allocated to treating patients. For radiology specifically, this could include significant changes in nonimaging aspects, such as patient scheduling, insurance and billing information, and managing insurance claims. AI’s streamlining of laborious administrative tasks will undoubtedly require radiology scheduling and billing departments to restructure in a way that maximizes the strengths of both human and AI capabilities, ultimately benefiting the patient experience. By automating many of the nonclinical aspects, radiologists and staff can focus more on patient care, improving the well-being of radiologists and reducing the risk of burnout, which will, in turn, lead to better overall outcomes.
From Automation to Innovation: How Artificial Intelligence Is Reshaping Global Industries.
Smith J, Fishman EK, Chu LC, Rowe SP, Crawford CK.
J Am Coll Radiol. 2025 Jun 19:S1546-1440(25)00340-0.
Adrenal
- For over 20 years, the two key tenets of adrenal incidentaloma (AI) evaluation have been the upper threshold of 10 Hounsfield units (HU) on noncontrast CT (ncCT) to delineate benignity, and the utilisation of adrenal washout CT (AWCT) to evaluate those above this cutoff. In light of growing recent evidence that challenges these two traditional principles, as well as re-evaluation of the data that led to their acceptance, we conclude that neither of these mainstays of adrenal CT remains relevant in modern AI diagnostic workup. With an appropriate definition of an incidentaloma and endocrine assessment for the majority of adrenal lesions, our analysis establishes that the use of AWCT should be ceased in the assessment of AIs, and that a 20 HU attenuation threshold for lesions < 4 cm should replace the traditional 10 HU threshold to exclude malignancy in this patient population. We therefore propose new recommendations for the management of AIs based primarily on CT attenuation and lesion size on ncCT.
Washed up: the end of an era for adrenal incidentaloma CT
James H. Seow et al.
Insights into Imaging ( 2025) 16:136 - Critical relevance statement
Increasing the CT attenuation threshold to 20 HU for lesions < 4 cm and eliminating washout CT for true adrenal incidentalomas, together with recommendations for endocrine assessment, will significantly decrease the over-investigation of overwhelmingly benign adrenal lesions, whilst confidently excluding malignancy.
Key Points
● True incidentalomas exclude current or prior extra-adrenal malignancy and clinically suspected adrenal disease.
● Adrenal washout CT was never proven in the malignancy-sparse true incidentaloma population.
● Hormonal correlation in parallel with < 20 HU and < 4 cm thresholds of homogeneous lesions on noncontrast CT excludes malignancy.
Washed up: the end of an era for adrenal incidentaloma CT
James H. Seow et al.
Insights into Imaging ( 2025) 16:136 
- Accordingly, we propose that in true AIs, combined 20 HU and 4 cm cut-offs exclude malignancy, and replace the established 10 HU threshold. We also highlight that even in those > 20 HU but < 4 cm, or > 4 cm but < 20 HU, malignancy rates are also extremely low.
Washed up: the end of an era for adrenal incidentaloma CT
James H. Seow et al.
Insights into Imaging ( 2025) 16:136 - Radiology reports should include a recommendation to perform a hormonal evaluation for all remaining AIs, which will identify nearly all PCC, as well as most ACC. Additionally, endocrine correlation identifies subclinically functional adenomas (including MACS), where clinical management may take precedence. Recommending additional endocrinologist referral may depend on local preferences, as in some regions, to manage cost and access issues, endocrine testing can be performed by primary care physicians, with endocrinologist referral limited to those with abnormal results.
Washed up: the end of an era for adrenal incidentaloma CT
James H. Seow et al.
Insights into Imaging ( 2025) 16:136 - On ncCT , homogeneous AIs ≤ 20 HU and ≤ 4 cm, or as per existing guidelines, < 10 HU and any size, are considered benign (Category-1), with no imaging follow-up required. AIs which are > 20 HU and 1–4 cm, OR 10–20 HU but > 4 cm, are highly likely benign (Category-2), and therefore 6–12 month ncCT is currently suggested as supported by most AI guidelines, purely to identify growth or stability. Those > 20 HU AND > 4 cm are considered higher risk (Category-3), with multidisciplinary meeting or surgical referral recommended, albeit acknowledging that most will still be of benign aetiology. As discussed prior, a more cautious approach is advised if AIs are > 40 HU and/or > 6 cm.
Washed up: the end of an era for adrenal incidentaloma CT
James H. Seow et al.
Insights into Imaging ( 2025) 16:136 
Washed up: the end of an era for adrenal incidentaloma CT
James H. Seow et al.
Insights into Imaging ( 2025) 16:136- Whilst the 10 HU threshold and AWCT have been ingrained in the radiology mindset over the last 2–3 decades, we believe it is now a timely end of an era for both these tenets of AI imaging. First, there is now sufficient evidence that a 20 HU threshold in AIs < 4 cm can safely replace the prior 10 HU limit. Second, due to its inherent inaccuracy and the extremely low incidence of malignancy in true AIs, AWCT has no role in the evaluation of ALs in patients without prior/current malignancy or suspected adrenal disease. Instead, AIs can be managed with ncCT, in parallel with endocrine testing, which complementarily improves detection of benign(and rarely malignant) hormonally active lesions.
Washed up: the end of an era for adrenal incidentaloma CT
James H. Seow et al.
Insights into Imaging ( 2025) 16:136
Deep Learning
- Introduction Artificial intelligence (AI) is rapidly gaining importance in medicine.1 Recent findings, however, indicate potential concerns from the patients’ and the public’s perspective.2 So far, such research focused on attitudes toward medical AI tools3 and AI-generated medical advice.4 In contrast, little is known about the public perception of physicians themselves who use AI. This online study explored how statements on different types of AI use (diagnostic, therapeutic, and administrative) influence the public’s perception of respective physicians.
Results Participants included 1276 adults (680 [53.3%] women, 584 [45.8%] men, 7 [0.5%] nonbinary individuals, and 5 participants [0.4%] who preferred to not disclose their gender; mean [SD] age, 46.2 [15.6] years). In every AI condition, the portrayed physician was perceived as significantly less competent (control: 3.85 [95%CI, 3.75-3.94] points; administrative AI: 3.71 [95%CI, 3.61-3.80] points; diagnostic AI: 3.66 [95%CI, 3.56-3.76] points; therapeutic AI: 3.58 [95%CI, 3.48-3.68] points), less trustworthy (control: 3.88 [95%CI, 3.79-3.96] points; administrative AI: 3.66 [95%CI, 3.57-3.75] points; diagnostic AI: 3.62 [95%CI, 3.52-3.72] points; therapeutic AI: 3.61 [95%CI, 3.50- 3.71] points), and less empathic (control: 4.00 [95%CI, 3.92-4.09] points; administrative AI: 3.80 [95%CI, 3.71-3.88] points; diagnostic AI: 3.82 [95%CI, 3.73-3.92] points; therapeutic AI: 3.72 [95% CI, 3.62-3.82] points) compared with the control condition (Table and Figure). Moreover, participants indicated a significantly lower willingness to make an appointment with the portrayed physician, if any type of AI use was mentioned (control: 3.61 [95%CI, 3.50-3.73] points; administrative AI: 3.32 [95%CI, 3.21-3.44] points; diagnostic AI: 3.16 [95%CI, 3.03-3.30] points; therapeutic AI: 3.15 [95%CI, 3.01-3.29] points). There was no significant difference between the AI conditions for any rating dimension. - From the physician’s perspective it thus may be important to transparently communicate the rationale for using AI and to emphasize its potential benefits for the patient. Limitations to the The generalizability of our results are the use of hypothetical scenarios, the somewhat artificial nature ofour stimuli, and the recruitment of a sample that agreed to participate in such experiments. Future research should extend these findings to even more realistic settings and explore potential moderating factors, such as patients’ experience with AI and with digital tools in general.
Public Perception of Physicians Who Use Artificial Intelligence.
Reis M, Reis F, Kunde W.
JAMA Netw Open. 2025 Jul 1;8(7):e2521643. - IMPORTANCE Artificial intelligence (AI) presents transformative opportunities to address the increasing challenges faced by health care systems globally. Particularly, in data-rich environments, such as intensive care units (ICUs), AI could assist in enhancing clinical decision-making, streamline workflows, and improve patient outcomes. Despite these promising applications, the practical implementation of AI in clinical settings remains limited.
OBJECTIVE To systematically evaluate AI system operationalization in the ICU, focusing on the AI field’s progress over time, technical maturity, and risk of bias.
Operationalization of Artificial Intelligence Applications in the Intensive Care Unit: A Systematic Review.
Berkhout WEM, van Wijngaarden JJ, Workum JD, et al.
JAMA Netw Open. 2025 Jul 1;8(7):e2522866. doi: 10.1001/jamanetworkopen.2025.22866. PMID: 40699572 - CONCLUSIONS AND RELEVANCE Despite substantial growth in AI research within intensive care medicine in recent years, the transition from development to clinical implementation still remains limited and has made little progress over time. A paradigm shift is urgently required in the medical literature—one that moves beyond retrospective validation toward the operationalization and prospective testing of AI for tangible clinical impact.
Operationalization of Artificial Intelligence Applications in the Intensive Care Unit: A Systematic Review.
Berkhout WEM, van Wijngaarden JJ, Workum JD, et al.
JAMA Netw Open. 2025 Jul 1;8(7):e2522866. doi: 10.1001/jamanetworkopen.2025.22866. PMID: 40699572 - CONCLUSIONS AND RELEVANCE Our systematic review highlights a steep increase in AI research within intensive care medicine, largely driven by retrospective studies, but progress toward clinical implementation across all types of AI, including generative AI, remains limited. These findings suggest that a paradigm shift is urgently required to move toward the operationalization and prospective testing of AI applications to warrant their clinical benefit and impact. Living systematic AI reviews could provide an essential framework for tracking progress and identifying persistent gaps to expedite this transition.
Operationalization of Artificial Intelligence Applications in the Intensive Care Unit: A Systematic Review.
Berkhout WEM, van Wijngaarden JJ, Workum JD, et al.
JAMA Netw Open. 2025 Jul 1;8(7):e2522866. doi: 10.1001/jamanetworkopen.2025.22866. PMID: 40699572
- Our analysis encompassed all document types available in Web of Science, including articles, published conference contributions, review articles, and editorial materials. We found that the number of articles authored by researchers affiliated with Chinese institutions has steadily increased, reaching a total of 108,282 articles published in English and 3489 in Chinese. In fact, annual publication numbers surpassed those in the United States in 2022 and 2023. The average annual growth rate of publications in China was 21.0%, exceeding the growth rate in the United States (9.9%).
The Landscape of Medical AI in China
Yue Qiu , Weizhi Ma , Haibo Wang et al.
NEJM AI 2025;2(7) - We classified publications into three broad groups: technical development (e.g., articles on the development of AI algorithms and large language models); clinical disciplines (e.g., articles on the application of AI for disease diagnosis and management); and management (e.g., policy articles). Technical development articles (72.1%) formed the majority of China’s publications, followed by clinical applications (21.5%). In contrast, the United States had a distribution of 51.4% in technical fields, 30.8% in clinical fields, and 6.1% in management fields.
The Landscape of Medical AI in China
Yue Qiu , Weizhi Ma , Haibo Wang et al.
NEJM AI 2025;2(7) 
- The regulatory framework may collectively foster an improved ecosystem for medical AI development in China. Nevertheless, significant challenges persist, including data fragmentation, regulatory and security concerns, and the difficulty of integrating AI technologies into clinical practice and hospital workflow. Addressing these barriers will be essential for China and other countries to fully realize the potential of medical AI.
The Landscape of Medical AI in China
Yue Qiu , Weizhi Ma , Haibo Wang et al.
NEJM AI 2025;2(7) - China’s medical AI landscape has progressed rapidly in recent years, with substantial growth in both the number and quality of research publications, including contributions from both public-sector academic institutions and private-sector companies. Despite this progress, the impact of China’s medical AI development still lags behind that of leading institutions in the United States. Progress in research funding, scaling computational infrastructure, enabling data sharing and utilization, and enhancing the regulatory framework may collectively foster an improved ecosystem for medical AI development in China.
The Landscape of Medical AI in China
Yue Qiu , Weizhi Ma , Haibo Wang et al.
NEJM AI 2025;2(7)
- Results The sensitivity, specificity, PPV, and NPV for tumor mass detection were 77.0%, 76.0%, 75.5%, and 77.5%, respectively; for D/P ratio detection, 87.0%, 94.2%, 93.5%, and 88.3%, respectively; and for combined tumor mass and D/P ratio detections, 96.0%, 70.2%, 75.6%, and 94.8%, respectively. No significant difference was observed between the software’s sensitivity and that of the radiologist’s report (software, 96.0%; radiologist, 96.0%; p = 1). The concordance rate between software findings and EUS was 96.0%. Conclusions Combining indirect indicator evaluation with tumor mass detection may improve small PDAC detection accuracy.
Deep learning-based automatic detection of pancreatic ductal adenocarcinoma ≤ 2 cm with high-resolution computed tomography: impact of the combination of tumor mass detection and indirect indicator evaluation.
Ozawa M, Sone M, Hijioka S, et al.
Jpn J Radiol. 2025 Jul 18. doi: 10.1007/s11604-025-01836-z. Epub ahead of print. PMID: 40679757. - High-resolution contrast-enhanced computed tomography (CT) scans from 181 patients diagnosed with PDAC (diameter ≤ 2 cm) between January 2018 and December 2023 were analyzed. The D/P ratio, which is the cross-sectional area of the MPD to that of the pancreatic parenchyma, was identified as an indirect indicator. A total of 204 patient data sets including 104 normal controls were analyzed for automatic tumor mass detection and D/P ratio evaluation. The sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were evaluated to detect tumor mass. The sensitivity of PDAC detection was compared with that of the software and radiologists, and tumor localization accuracy was validated against endoscopic ultrasonography (EUS) findings.
Deep learning-based automatic detection of pancreatic ductal adenocarcinoma ≤ 2 cm with high-resolution computed tomography: impact of the combination of tumor mass detection and indirect indicator evaluation.
Ozawa M, Sone M, Hijioka S, et al.
Jpn J Radiol. 2025 Jul 18. doi: 10.1007/s11604-025-01836-z. Epub ahead of print. PMID: 406777. - Small PDACs are often challenging to detect as distinct tumor masses. Consequently, indirect indicators, such as dilatation of the main pancreatic duct (MPD) and atrophy of the pancreatic parenchyma, may play a crucial in diagnosis. For example, compared with the detection rates of 51.5% for tumor masses of small PDAC, that for MPD dilatation is relatively high at 79.6%. Thus, we hypothesized that combining indirect indicators and automatic tumor mass detection could improve the diagnostic accuracy of small PDAC. Furthermore, high-resolution CT (HR-CT), known for its superior low-contrast detectability, may further improve the diagnosis of small PDAC.
Deep learning-based automatic detection of pancreatic ductal adenocarcinoma ≤ 2 cm with high-resolution computed tomography: impact of the combination of tumor mass detection and indirect indicator evaluation.
Ozawa M, Sone M, Hijioka S, et al.
Jpn J Radiol. 2025 Jul 18. doi: 10.1007/s11604-025-01836-z. Epub ahead of print. PMID: 406777. - “Indirect indicators such as MPD dilatation play a vital role in identifying small PDAC, and their clinical importance has been previously reported. A retrospective observational study demonstrated that 88% of isoattenuating small PDAC cases present are indirect indicators. Another multicenter retrospective observational study reported that MPD dilatation and focal fatty changes in the pancreatic parenchyma were observed in 79.6% and 41.8% of patients with small PDAC, respectively.”
Deep learning-based automatic detection of pancreatic ductal adenocarcinoma ≤ 2 cm with high-resolution computed tomography: impact of the combination of tumor mass detection and indirect indicator evaluation.
Ozawa M, Sone M, Hijioka S, et al.
Jpn J Radiol. 2025 Jul 18. doi: 10.1007/s11604-025-01836-z. Epub ahead of print. PMID: 406777.
- Radiomics is a mathematical approach to medical images to extract quantitative features , generating a “radiomics signature.” The radiomics workflow involves image acquisition and pre-processing, region of interest segmentation, feature extraction, and model training and validation. It has generated promising results; however, clinical implementation for early detection remains a challenge. Pancreatic ductal adenocarcinoma (PDAC), the most common pancreatic cancer, has a highly aggressive nature with an aggregated 5-year survival rate of only 13%. Early detection of PDAC provides timely surgical intervention, with the hope of improved survival rates. Radiomics has been applied to the detection of PDAC; however, its sensitivity to variations in image acquisition parameters has posed significant challenges, limiting the development of robust and generalizable models. This review explores the current landscape of radiomics for the early detection of PDAC, highlighting key challenges within the radiomics workflow and barriers to its progression from a proof-of-concept into clinical practice.
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579. - Since surgical resection is considered as the only potential “cure” for PDAC, early detection at a resectable stage has hopes for improving overall survival. A comprehensive strategy for the early identification of PDAC involves focused screening for high-risk individuals, the use of serum biomarkers, liquid biopsies, and AI-assisted tumor detection in radiologic imaging. With advancement in artificial intelligence (AI), radiomics has emerged as a widespread method of image analysis that could potentially aid in early detection and characterization of pancreatic cancer. Radiomics, a high throughput methodology that extracts quantitative mineable characteristics from radiologic images, is being widely explored to detect early pancreatic neoplastic changes, often imperceptible by the human eye.
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579. - Radiomics is a mathematical approach to medical imaging, which aims to extract quantitative information from diagnostic images. It allows the quantitative extraction of intensity- histogram statistics, texture features, and shape descriptors from a region of interest (ROI). Compared to visual assessment, automatic extraction of radiomics features reduces subjectivity in image interpretation.11-13 Radiomics is largely being explored in the field of oncology, as tumors are considered to have early subtle changes in imaging that can be imperceptible to the human eye.
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579. - The substantial spatial and temporal heterogeneity of solid tumors (ie, spanning genetic, proteomic, cellular, microenvironmental, tissue, and organ levels) limits the effectiveness of biopsy-based molecular assays. However, this complexity underscores the vast potential of advanced analytical techniques such as radiomics. It has also been shown that in addition to capturing tumor biological characteristics, radiomics signatures also reflect the underlying genomic and transcription profiles of the tumor, further enhancing its underlying potential.
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579. - Segmentation of the ROI can be manual or fully/semiautomated. Manual segmentation is highly labor-intensive and time consuming and variations occur among different segmentators. Automated segmentation algorithms have shown potential for entire organs, but tumor segmentation remains a challenge due to heterogenous tumor morphology among patients. This is particularly true for the pancreas because of its varying anatomy and indistinct borders with the surrounding tissues. Several novel deep learning (DL) models have been proposed for automated pancreatic segmentation on CT, MRI, and endoscopic ultrasound with Dice similarity coefficient (ie, Dice score) ranging from 0.6 to 0.96. TotalSegmentator, a widely known segmentation algorithm, achieves a Dice score of 0.7 for normal pancreas on both CT and MRI. Newer emerging models include VISTA 3D by NVIDIA with a Dice score of 0.82 for pancreas, however, it offers promptable automatic segmentation and interactive human editing which achieved the Dice scores of 0.638 for pancreatic tumors.
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579. - The Image Biomarker Standardization Initiative guidelines (IBSI) have standardized 169 radiomic features, 8 convolutional filters and 458 radiomic features from filtered images to enable reproducibility of radiomic studies.Radiomics features can be categorized into: (i) first-order features, corresponding to distribution of voxel intensities within the ROI; (ii) morphological features, corresponding to geometric aspects of the ROI, such as area and volume; and (iii) texture features, corresponding to high-order gray information within the ROI. Feature extraction is commonly performed using the PyRadiomics framework.
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579. - The selected features are used to train a machine learning (ML) model. Radiomics features have been leveraged to develop both supervised and unsupervised machine learning models. Supervised models are typically used for outcome prediction, whereas unsupervised analyses focus on clustering features and reveal underlying patterns within the dataset. Several classifier models including random forest, naïve Bayes classifier, and gradient boosting machines (eg, XGBoost) have been explored for differentiating between normal pancreatic tissue and PDAC.Other studies have utilized support vector machines (SVM) and ordinary least squares (OLS) logistic regression models to evaluate textural changes in pancreas in pre-diagnostic scans that could serve as indicators of PDAC development. Clinical characteristics,genomics, and proteomic data can be combined with radiomics features to generate highly efficient models that could aid in early detection, risk stratification, and survival prediction.
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579. - The first common use of radiomics analysis is to accurately differentiate between normal pancreatic tissue and PDAC. This would help in detecting small PDACs (<2 cm) early on CT scans, directing timely interventions. Chu et al,40 Chen et al,41 and Wang et al46 have demonstrated promising results in differentiating PDAC from normal pancreatic tissue using radiomic features. Chu et al analyzed manually segmented CT images and trained a random forest binary classification model A model to differentiate between PDAC and normal tissue. They found that a radiomics-based classification model achieved 100% sensitivity in the detection of PDAC. However, their study was limited by small sample size, variations in scanning protocols, and age mismatch between healthy controls and PDAC patients, which may have influenced radiomics-based classification due to age-related fatty and atrophic pancreatic changes.
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579. - Abdominal CT is the most frequently used imaging technique for different intraabdominal conditions. This results in large datasets and allows for a retrospective examination of “normal pre-diagnostic CT scans” in patients who subsequently developed a PDAC. Pre-diagnostic CT scans are defined as incidental CT scans performed for unrelated indications before PDAC diagnosis. Javed et al trained a naïve Bayes classification model employing radiomics to differentiate between high-risk and low-risk pancreas on pre-diagnostic CT scans. They achieved a classification accuracy of 93% on the internal dataset and concluded that early local textural changes, which can be detected by radiomics, occurred in the sub-region of the pancreas corresponding to tumor development later.
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579. - Mukherjee et al further validated the effectiveness of radiomics-based ML models and demonstrated that AI can detect PDAC at a median of 386 (range: 97-1092) days before clinical diagnosis by identifying subtle changes that are imperceptible to the human eye. They extracted 88 first order and gray-level texture features and evaluated 4 ML classifiers, with SVM emerging as the best performer with an AUC of 0.98 (95% CI: 0.94, 0.98) and accuracy of 92% (95% CI: 86.7, 93.7). Their AI models significantly outperformed radiologists, who exhibited only fair inter-reader agreement (Cohen’s kappa, 0.3) and a substantially lower diagnostic performance with an AUC of 0.66 (95% CI: 0.46, 0.86).
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579. - “While radiomics possesses significant potential, there are several drawbacks that hinder its practical application. Developing a radiomic pipeline necessitates the processing of large imaging datasets, which in turn demands substantial computational resources, efficient data storage and management systems, and skilled personnel. A dedicated multidisciplinary team, comprising radiologists, data scientists, and engineers, is essential for the development and validation of the radiomic workflow. This extensive workflow is both time-consuming and resource intensive, heavily dependent on the expertise of the production team. There is a need for user-independent programs that clinicians can operate with ease.”
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579. - Many radiomics studies do not justify their sample size, leading to models that may be too small to avoid overfitting, limiting their widespread adoption and reproducibility. These models may yield very high accuracy on training datasets and perform poorly on external validation sets. There is a need for large, multi-center studies to bridge this validation gap, however, data sharing among institutions remains a critical barrier. Ahmed et al assessed the radiomics quality score across 54 studies focused on pancreatic imaging and reported a low mean score of 32.1%, reflecting significant methodological heterogeneity and limited reproducibility of the current Radiomics models. Notably, none of the studies included a cost-effectiveness analysis, raising concerns about clinical integration of these models. Additionally, radiomics-based Studies have shown low adherence to reporting guidelines and quality frameworks, including IBSI and CLAIM (Checklist for Artificial Intelligence in Medical Imaging), further limiting The study's reproducibility.
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579.- The challenges are multifaceted, stemming from the complexities involved in constructing and validating a radiomics model. The radiomics pipeline is highly sensitive to data variations introduced by differing scan protocols and scanner manufacturers, limiting the construction of a robust reproducible radiomics model. Furthermore, leading vendors such as Siemens Healthineers, GE Healthcare, Phillips, Canon, and United Imaging offer a range of 5 to 10 different imaging systems, each utilizing distinct algorithms, which adds to the complexity of the issue. Even with the same vendors, institutions tend to have institution-specific protocols that contribute to additional variability. To address these challenges, future efforts could focus on the development of parallel image reconstruction pipelines, one already existing optimized for human evaluation and another specifically tailored to support AI model development.
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579. - Radiomics has revolutionized the paradigm of early detection of PDAC with promising results but remains largely unused in clinical practice. Bridging the gap to clinical implementation will require overcoming challenges in both the radiomics workflow and the clinical side. Standardized imaging protocols, optimized image filters, and reconstruction parameters tailored for radiomics could help develop robust and reproducible models. However, the binary output of the model, which is dependent on the training set, restricts its ability to detect different pancreatic lesions, thus limiting its real-world applicability. Due to the overall low incidence of PDAC, routine screening is not recommended, but radiomics combined with multi-cancer early detection may assist in early detection using imaging data obtained for other causes. While this is an interesting concept to explore, the generation of a high number of false positives remains a significant concern.
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579. 
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Kidney
- Results The study group comprised of 134 patients (90 males) with 265 lesions (229 RCC-Mets and 36 PNETs). Patients with PNETs were significantly younger (62 ± 12 years vs. 67 ± 9 years, p = 0.013). Inter-observer agreement for CT/MRI features was excellent across multiple imaging variables (k = 0.86–1.00). Most PNETs were single lesions (88 vs. 63%, p = 0.008), smaller in size (14 mm vs. 23 mm, p = 0.042), more common in the body and tail (81 vs. 57%, p = 0.01), showedhomogeneous contrast enhancement (64–79% vs. 39–49%, p < 0.01–0.03), less T1-hypointense (80 vs. 99%, p = 0.002) and more DWI hyperintense (71 vs. 58%, p < 0.001) compared to RCC-Mets.
Conclusion PNETs are typically single, occur in distal pancreas, and enhance homogeneously compared to RCC-Mets which are often multiple, occur in the proximal pancreas, and enhance heterogeneously.
Differentiation between renal cell carcinoma metastases to the pancreas and pancreatic neuroendocrine tumors in patients with renal cell carcinoma on CT or MRI.
Nduwimana MJ, et al.
Abdom Radiol (NY). 2025 Jul;50(7):3115-3125 - Purpose To determine whether renal cell carcinoma metastases (RCC-Mets) to the pancreas can be differentiated from pancreatic neuroendocrine tumors (PNETs) in patients with RCC on CT or MRI at presentation
Conclusion PNETs are typically single, occur in distal pancreas, and enhance homogeneously compared to RCC-Mets which are often multiple, occur in the proximal pancreas, and enhance heterogeneously.
Differentiation between renal cell carcinoma metastases to the pancreas and pancreatic neuroendocrine tumors in patients with renal cell carcinoma on CT or MRI.
Nduwimana MJ, Colak C, Bilgin C, et al.
Abdom Radiol (NY). 2025 Jul;50(7):3115-3125.. - Metastases to the pancreas are rare with an estimated incidence of 2–5% in clinical series . Common malignancies that metastasize to the pancreas include renal cell carcinoma (RCC), lung carcinomas, breast carcinoma, colorectal carcinoma, melanoma and soft-tissue sarcomas . RCC is the most common primary neoplasm to metastasize to the pancreas, accounting for approximately 30% of pancreatic metastases . RCC is also the most common malignancy to present with isolated pancreatic metastases. Though rare, pancreatic neuroendocrine tumors (PNETs) are the second most common primary pancreatic neoplasm with varying biologic behavior from benign to malignant.
Differentiation between renal cell carcinoma metastases to the pancreas and pancreatic neuroendocrine tumors in patients with renal cell carcinoma on CT or MRI.
Nduwimana MJ, Colak C, Bilgin C, et al.
Abdom Radiol (NY). 2025 Jul;50(7):3115-3125.. - PNETs more commonly developed in younger patients (62 ± 12 years vs. 67 ± 9 years, p = 0.013), in males [27/32 (84.4%) vs. 63/102 (61.8%), p = 0.017], and were diagnosed earlier following nephrectomy (5.3 ± 9.7 years vs. 9.9 ± 7.4 years, p = 0.007) compared to the RCC-Mets group (Table 1). The number of pancreatic lesions ranged from 1 to 27, with a mean ± SD of 2 ± 2.9 per patient. There was no significant difference in the number of lesions per patient between RCC-Mets and PNETs (2.2 ± 3.2 vs. 1.2 ± 0.4, p = 0.062). However, PNETs more commonly presented as single lesions (28/32 = 88% vs. 64/102 = 63%, p = 0.008), and were more frequently located in the distal pancreas (80.6 vs. 57.2%, p = 0.01), compared to RCC-Mets,
Differentiation between renal cell carcinoma metastases to the pancreas and pancreatic neuroendocrine tumors in patients with renal cell carcinoma on CT or MRI.
Nduwimana MJ, Colak C, Bilgin C, et al.
Abdom Radiol (NY). 2025 Jul;50(7):3115-3125.. - In our study, though most of the PNETs were histopathologically proven grade 1, nearly half of the cases underwent,resection. However, current management for PNETs suggests that surgery be driven by tumor size (> 2 cm) and/ or poor histopathologic differentiation (grade 3) as the disease is mostly benign. In contrast, most RCC-Mets underwent pathologic confirmation through biopsy and only 54% of patients underwent surgical resection, despite it being the primary treatment for the disease. Resection is linked to favorable prognosis and extended survival . Moreover, pancreatic biopsy is a challenging and invasive procedure due to the pancreas’ anatomic location. The above findings may be attributed to the two-decade span of our study, during which treatment guidelines have evolved. Accurately distinguishing RCC-Mets from PNETs on imaging could facilitate earlier surgical intervention and potentially improving survival outcomes. Additionally, with RCCMets presenting even up to two decades after nephrectomy, routine, long-term, radiologic surveillance is encouraged.
Differentiation between renal cell carcinoma metastases to the pancreas and pancreatic neuroendocrine tumors in patients with renal cell carcinoma on CT or MRI.
Nduwimana MJ, Colak C, Bilgin C, et al.
Abdom Radiol (NY). 2025 Jul;50(7):3115-3125.. - Lastly, several different hypervascular lesions can occur within and around pancreas that need to be differentiated from RCC-Mets and PNETs, including intrapancreatic accessory spleen, solid pseudopapillary tumor, solid serous cystadenoma, acinar cell carcinoma, hypervascular metastases, and gastrointestinal stromal tumor (GIST). Intrapancreatic accessory spleens follow the enhancement pattern of the spleen on all contrast enhanced phases and have similar signal intensity as spleen parenchyma on MRI. Solid pseudopapillary tumors almost exclusively occurs in young females and may present with heterogeneous enhancement and T1 hyperintensity due to hemorrhage. Solid appearing serous cystadenomas typically occur in asymptomatic elderly woman and may rarely appear solid; however, characteristic small cystic spaces may be visible on MRI differentiating from other lesions.
Differentiation between renal cell carcinoma metastases to the pancreas and pancreatic neuroendocrine tumors in patients with renal cell carcinoma on CT or MRI.
Nduwimana MJ, Colak C, Bilgin C, et al.
Abdom Radiol (NY). 2025 Jul;50(7):3115-3125.. - In conclusion, there are significant differences between RCC-Mets and PNET lesions in patients with RCC on CT or MRI. PNETs are usually single, seen in the body and tail, and exhibit homogeneous enhancement. Whereas RCC-Mets to the pancreas are more often multiple, seen in the head, uncinate and neck region of the pancreas, and exhibit heterogeneous enhancement. These imaging differences may be useful for early diagnosis to triage appropriate follow-up management.
Differentiation between renal cell carcinoma metastases to the pancreas and pancreatic neuroendocrine tumors in patients with renal cell carcinoma on CT or MRI.
Nduwimana MJ, Colak C, Bilgin C, et al.
Abdom Radiol (NY). 2025 Jul;50(7):3115-3125..
Musculoskeletal
- Cellulitis
Uncomplicated cellulitis is a clinical diagnosis and is treated conservatively with antibiotics and locally supportive measures. However, there is a risk for serious and rapid spread of the infection in patients with diabetes, immunodeficiency, impaired peripheral circulation or history of lymphadenectomy. For such patients in the ER, CT is used to accurately differentiate between superficial cellulitis and cellulitis associated with a deep-seated infection. In uncomplicated cellulitis, CT demonstrates skin thickening, stranding and septation of the subcutaneous fat and thickening of the underlying superficial fascia. A foreign body associated with cellulitis can be detected by CT, if present . - Necrotizing Fasciitis
Necrotizing fasciitis is a progressive, rapidly spreading infection located in the deep fascia, with secondary necrosis of the subcutaneous tissues . The speed of spread of the infection is directly proportional to the thickness of the subcutaneous layer. The occurrence of necrotizing fasciitis is relatively rare, though it is on the rise because of an increase in immunocompromised patients with HIV, diabetes mellitus, cancer, alcoholism, vascular insufficiencies and organ transplantation. It can occur after trauma or around foreign bodies in surgical wounds, though it can also be idiopathic, as in scrotal or penile necrotizing fasciitis. The majority of necrotizing soft tissue infections have gas-forming anaerobic bacteria present, usually in combination with aerobic gram-negative organisms. - Necrotizing Fasciitis
Necrotizing fasciitis is a life-threatening surgical emergency. Unfortunately, this infection can be difficult to recognize in its early stages, but it rapidly progresses. The overall morbidity and mortality is 70-80%, and one of the most important predictors of mortality is a delay in the diagnosis of necrosis. Hence, CT imaging can play a vital role in suggesting the diagnosis early and initiating rapid and successful treatment. - Necrotizing Fasciitis
The imaging findings in necrotizing fasciitis are similar to those as with cellulitis, but are more severe and show involvement of deeper structures. One specific distinguishing sign of necrotizing fasciitis is the occurrence of gas in the subcutaneous tissues due to the presence of gas-forming anaerobic organisms (Fig. 6). However, it should be noted that gas is not observed with all cases of necrotizing fasciitis. Other CT features include thickening of the affected fascia, fluid collections along the deep fascial sheaths as well as extension of edema into the intermuscular septae and the muscles (Fig. 7, 8). Following contrast enhancement, there is no demonstrable enhancement of the fascia, confirming the presence of necrosis and distinguishing non-necrotizing fasciitis from necrotizing fasciitis. Those patients with non-necrotizing fasciitis are not surgical emergencies, but they should be followed for the potential of forming necrosis. - Soft Tissue Abscess
Although most bacterial infections in the soft tissues stay localized, a soft tissue abscess may form, particularly in immunocompromised patients. The most common isolated pathogen is Staphylococcus Aureus, though in an urban emergency department population, MRSA is now cultured in 51% of patients with a soft tissue infection and such patients are more likely to present with a soft tissue abscess than patients from whom other bacteria are cultured. By CT, a well-demarcated fluid collection with a peripheral pseudocapsule showing rim enhancement is characteristic of an abscess and differentiates an abscess from simple cellulitis or fasciitis. The treatment of a soft tissue abscess is administration of appropriate antibiotics and percutaneous drainage. - Infectious Myositis
Infectious myositis is an acute, subacute or chronic infection of skeletal muscle that is most commonly seen in young adults. While viruses, bacteria (including mycobacteria), fungi, and parasites can all cause myositis, the most common infectious agent is a bacteria, namely Staphylococcus aureus, responsible for over 75% of cases. Pyomyositis or bacterial myositis, was once considered a tropical disease, but is now seen in temperate climates, particularly with the emergence of HIV infection, where, in one study, 17% of patients with pyomyositis had underlying HIV infection. In fact, some authors report pyomyositis as the most common musculoskeletal complication of AIDS, but other risk factors abound and include strenuous activity or rhabdomyolysis and a history of muscle trauma where a hematoma may form and act as a nidus for infection. Skin infections, infected insect bites, illicit drug injections, and underlying diabetes mellitus can also lead to pyomyositis. - Osteomyelitis
Osteomyelitis is an infection of the bone that can result from hematogenous spread or be secondary to direct or contiguous inoculation. In young adults, it is most commonly associated with an open fracture or direct trauma while in elderly and pediatric patients, the cause of osteomyelitis is typically bacteremia. As with other musculoskeletal infections, disease states known to predispose patients to osteomyelitis include immunosuppression, diabetes mellitus, sickle cell disease, intravenous drug abuse and alcoholism. - Osteomyelitis
Hematogenous osteomyelitis usually presents with a slow insidious progression of symptoms, while osteomyelitis due to direct inoculation is localized, with prominent local signs and symptoms. The most frequently involved bones are the tibia, wrist, femur, ribs and thoracolumbar spine. The most common pathogen cultured is Staphylococcus Aureus, although in the HIV population, 30% of the cases of osteomyelitis are due to atypical mycobacteria. It should be noted that blood culture results are positive in only 50% of patients with hematogenous osteomyelitis. - Osteomyelitis
By CT, features of bacterial osteomyelitis include overlying soft tissue swelling, periosteal reaction, medullary lucencies or trabecular coursening and focal cortical erosions. In addition, the observation of an extramedullary fat-fluid sign is a rare but specific sign for osteomyelitis. This sign is an indication of cortical breach and, thus, in the absence of trauma, confirms the presence of osteomyelitis. For chronic osteoomyelitis, CT is considered superior to MRI for the demonstration of cortical destruction, gas and sequestra. - Osteomyelitis
As stated, the most common pathogen associated with osteomyelitis is Staphylococcus Aureus. However, mycobacterial infections are of great concern in the ER of a large center city hospital. The patient group at greatest risk to develop tuberculosis is the population infected with HIV. In long bones, tuberculous osteomyelitis usually begins in the metaphysis and spreads to the epiphysis. Eventually, cortical erosion into the joint occurs. Typically, skeletal tuberculosis presents with distinct bony margins without evidence of periosteal reaction (unless occurring in children). However, the most common osseous site of disease is the spine, comprising 50% of cases, and in particular, tuberculosis affects the thoracic spine. Early infection findings by CT include vertebral osteopenia followed later by slight disc space narrowing and then, characteristic anterior corner bone destruction. Skip lesions can occur as well as spread to the posterior elements, a rare but specific sign for tuberculosis. Paravertebral soft tissue edema and abscess formation occurs and chronically, calcification in the wall of these abscesses can be detected by CT .
- Neurofibromas are tumors derived from Schwann cells, fibroblasts, and supporting cells known as perineural cells. Typically, they are benign and manifest as multiple tumors. NF1 is an autosomal dominant genetic disorder with a prevalence of approximately 1 in 4,000 births and no racial predilection. NF1 is characterized by multiple neurofibromas along the peripheral nerves, optic nerve gliomas, sphenoid wing dysplasia, pigmented iris nodules, and hyperpigmented macular skin lesions known as café-au-lait spots. It is associated with a gene on chromosome 17. The formation of dermal neurofibromas is a hallmark of NF1 with a characteristic distribution on the trunk and sparing of the extremities . With time, neurofibromas may undergo malignant degeneration.
Malignant peripheral nerve sheath tumor.
Hrehorovich PA, Franke HR, Maximin S, Caracta P.
Radiographics. 2003 May-Jun;23(3):790-4. - MPNSTs most commonly occur in the deep soft tissues, usually close to a nerve trunk. The most common sites are the sciatic nerve, brachial plexus, and sacral plexus. The past literature referred to MPNST as malignant schwannoma, neurogenic sarcoma, and neurofibrosarcoma. Malignant peripheral nerve sheath tumor is the current term used by the World Health Organization for this highly aggressive tumor. MPNSTs may arise from plexiform neurofibromas, de novo or secondary to radiation therapy. At histologic analysis, the presence of mitotic figures distinguishes MPNST from otherwise typical neurofibromas. MPNST accounts for approximately 10% of soft-tissue sarcomas, and 40%–60% of MPNSTs arise from cases of NF1. Overall, there is a 4% chance of malignant transformation in NF1.
Malignant peripheral nerve sheath tumor.
Hrehorovich PA, Franke HR, Maximin S, Caracta P.
Radiographics. 2003 May-Jun;23(3):790-4. - Clinically, pain is a classic presenting symptom in patients with MPNST. Other findings include masses larger than 2–6 cm with irregular borders and a history of rapid growth. Often, MPNST produces neurologic deficits in the distribution of the involved nerves due to impingement or mass effect (,10,,11). At CT, the attenuation of neurogenic tumors depends on their histologic characteristics. Neurofibromas typically have low attenuation, which may be related to the fat content of myelin from Schwann cells, the high water content of myxoid tissue, entrapment of fat, and cystic areas of hemorrhage and necrosis (,5). Central enhancement or a target appearance may be seen due to the less cellular and vascular myxoid tissue located in the periphery and the more vascular fibrous tissue seen centrally.
Malignant peripheral nerve sheath tumor.
Hrehorovich PA, Franke HR, Maximin S, Caracta P.
Radiographics. 2003 May-Jun;23(3):790-4. - At CT, the attenuation of neurogenic tumors depends on their histologic characteristics. Neurofibromas typically have low attenuation, which may be related to the fat content of myelin from Schwann cells, the high water content of myxoid tissue, entrapment of fat, and cystic areas of hemorrhage and necrosis . Central enhancement or a target appearance may be seen due to the less cellular and vascular myxoid tissue located in the periphery and the more vascular fibrous tissue seen centrally. The characteristic dumbbell lesion, a partly intradural and partly extradural tumor, represents a neurofibroma that expands the intervertebral foramina and may be best appreciated with cross-sectional imaging.
Malignant peripheral nerve sheath tumor.
Hrehorovich PA, Franke HR, Maximin S, Caracta P.
Radiographics. 2003 May-Jun;23(3):790-4.
Pancreas
- Results The sensitivity, specificity, PPV, and NPV for tumor mass detection were 77.0%, 76.0%, 75.5%, and 77.5%, respectively; for D/P ratio detection, 87.0%, 94.2%, 93.5%, and 88.3%, respectively; and for combined tumor mass and D/P ratio detections, 96.0%, 70.2%, 75.6%, and 94.8%, respectively. No significant difference was observed between the software’s sensitivity and that of the radiologist’s report (software, 96.0%; radiologist, 96.0%; p = 1). The concordance rate between software findings and EUS was 96.0%. Conclusions Combining indirect indicator evaluation with tumor mass detection may improve small PDAC detection accuracy.
Deep learning-based automatic detection of pancreatic ductal adenocarcinoma ≤ 2 cm with high-resolution computed tomography: impact of the combination of tumor mass detection and indirect indicator evaluation.
Ozawa M, Sone M, Hijioka S, et al.
Jpn J Radiol. 2025 Jul 18. doi: 10.1007/s11604-025-01836-z. Epub ahead of print. PMID: 40679757. - High-resolution contrast-enhanced computed tomography (CT) scans from 181 patients diagnosed with PDAC (diameter ≤ 2 cm) between January 2018 and December 2023 were analyzed. The D/P ratio, which is the cross-sectional area of the MPD to that of the pancreatic parenchyma, was identified as an indirect indicator. A total of 204 patient data sets including 104 normal controls were analyzed for automatic tumor mass detection and D/P ratio evaluation. The sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were evaluated to detect tumor mass. The sensitivity of PDAC detection was compared with that of the software and radiologists, and tumor localization accuracy was validated against endoscopic ultrasonography (EUS) findings.
Deep learning-based automatic detection of pancreatic ductal adenocarcinoma ≤ 2 cm with high-resolution computed tomography: impact of the combination of tumor mass detection and indirect indicator evaluation.
Ozawa M, Sone M, Hijioka S, et al.
Jpn J Radiol. 2025 Jul 18. doi: 10.1007/s11604-025-01836-z. Epub ahead of print. PMID: 406777. - Small PDACs are often challenging to detect as distinct tumor masses. Consequently, indirect indicators, such as dilatation of the main pancreatic duct (MPD) and atrophy of the pancreatic parenchyma, may play a crucial in diagnosis. For example, compared with the detection rates of 51.5% for tumor masses of small PDAC, that for MPD dilatation is relatively high at 79.6%. Thus, we hypothesized that combining indirect indicators and automatic tumor mass detection could improve the diagnostic accuracy of small PDAC. Furthermore, high-resolution CT (HR-CT), known for its superior low-contrast detectability, may further improve the diagnosis of small PDAC.
Deep learning-based automatic detection of pancreatic ductal adenocarcinoma ≤ 2 cm with high-resolution computed tomography: impact of the combination of tumor mass detection and indirect indicator evaluation.
Ozawa M, Sone M, Hijioka S, et al.
Jpn J Radiol. 2025 Jul 18. doi: 10.1007/s11604-025-01836-z. Epub ahead of print. PMID: 406777. - “Indirect indicators such as MPD dilatation play a vital role in identifying small PDAC, and their clinical importance has been previously reported. A retrospective observational study demonstrated that 88% of isoattenuating small PDAC cases present are indirect indicators. Another multicenter retrospective observational study reported that MPD dilatation and focal fatty changes in the pancreatic parenchyma were observed in 79.6% and 41.8% of patients with small PDAC, respectively.”
Deep learning-based automatic detection of pancreatic ductal adenocarcinoma ≤ 2 cm with high-resolution computed tomography: impact of the combination of tumor mass detection and indirect indicator evaluation.
Ozawa M, Sone M, Hijioka S, et al.
Jpn J Radiol. 2025 Jul 18. doi: 10.1007/s11604-025-01836-z. Epub ahead of print. PMID: 406777.
- Radiomics is a mathematical approach to medical images to extract quantitative features , generating a “radiomics signature.” The radiomics workflow involves image acquisition and pre-processing, region of interest segmentation, feature extraction, and model training and validation. It has generated promising results; however, clinical implementation for early detection remains a challenge. Pancreatic ductal adenocarcinoma (PDAC), the most common pancreatic cancer, has a highly aggressive nature with an aggregated 5-year survival rate of only 13%. Early detection of PDAC provides timely surgical intervention, with the hope of improved survival rates. Radiomics has been applied to the detection of PDAC; however, its sensitivity to variations in image acquisition parameters has posed significant challenges, limiting the development of robust and generalizable models. This review explores the current landscape of radiomics for the early detection of PDAC, highlighting key challenges within the radiomics workflow and barriers to its progression from a proof-of-concept into clinical practice.
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579. - Since surgical resection is considered as the only potential “cure” for PDAC, early detection at a resectable stage has hopes for improving overall survival. A comprehensive strategy for the early identification of PDAC involves focused screening for high-risk individuals, the use of serum biomarkers, liquid biopsies, and AI-assisted tumor detection in radiologic imaging. With advancement in artificial intelligence (AI), radiomics has emerged as a widespread method of image analysis that could potentially aid in early detection and characterization of pancreatic cancer. Radiomics, a high throughput methodology that extracts quantitative mineable characteristics from radiologic images, is being widely explored to detect early pancreatic neoplastic changes, often imperceptible by the human eye.
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579. - Radiomics is a mathematical approach to medical imaging, which aims to extract quantitative information from diagnostic images. It allows the quantitative extraction of intensity- histogram statistics, texture features, and shape descriptors from a region of interest (ROI). Compared to visual assessment, automatic extraction of radiomics features reduces subjectivity in image interpretation.11-13 Radiomics is largely being explored in the field of oncology, as tumors are considered to have early subtle changes in imaging that can be imperceptible to the human eye.
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579. - The substantial spatial and temporal heterogeneity of solid tumors (ie, spanning genetic, proteomic, cellular, microenvironmental, tissue, and organ levels) limits the effectiveness of biopsy-based molecular assays. However, this complexity underscores the vast potential of advanced analytical techniques such as radiomics. It has also been shown that in addition to capturing tumor biological characteristics, radiomics signatures also reflect the underlying genomic and transcription profiles of the tumor, further enhancing its underlying potential.
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579. - Segmentation of the ROI can be manual or fully/semiautomated. Manual segmentation is highly labor-intensive and time consuming and variations occur among different segmentators. Automated segmentation algorithms have shown potential for entire organs, but tumor segmentation remains a challenge due to heterogenous tumor morphology among patients. This is particularly true for the pancreas because of its varying anatomy and indistinct borders with the surrounding tissues. Several novel deep learning (DL) models have been proposed for automated pancreatic segmentation on CT, MRI, and endoscopic ultrasound with Dice similarity coefficient (ie, Dice score) ranging from 0.6 to 0.96. TotalSegmentator, a widely known segmentation algorithm, achieves a Dice score of 0.7 for normal pancreas on both CT and MRI. Newer emerging models include VISTA 3D by NVIDIA with a Dice score of 0.82 for pancreas, however, it offers promptable automatic segmentation and interactive human editing which achieved the Dice scores of 0.638 for pancreatic tumors.
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579. - The Image Biomarker Standardization Initiative guidelines (IBSI) have standardized 169 radiomic features, 8 convolutional filters and 458 radiomic features from filtered images to enable reproducibility of radiomic studies.Radiomics features can be categorized into: (i) first-order features, corresponding to distribution of voxel intensities within the ROI; (ii) morphological features, corresponding to geometric aspects of the ROI, such as area and volume; and (iii) texture features, corresponding to high-order gray information within the ROI. Feature extraction is commonly performed using the PyRadiomics framework.
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579. - The selected features are used to train a machine learning (ML) model. Radiomics features have been leveraged to develop both supervised and unsupervised machine learning models. Supervised models are typically used for outcome prediction, whereas unsupervised analyses focus on clustering features and reveal underlying patterns within the dataset. Several classifier models including random forest, naïve Bayes classifier, and gradient boosting machines (eg, XGBoost) have been explored for differentiating between normal pancreatic tissue and PDAC.Other studies have utilized support vector machines (SVM) and ordinary least squares (OLS) logistic regression models to evaluate textural changes in pancreas in pre-diagnostic scans that could serve as indicators of PDAC development. Clinical characteristics,genomics, and proteomic data can be combined with radiomics features to generate highly efficient models that could aid in early detection, risk stratification, and survival prediction.
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579. - The first common use of radiomics analysis is to accurately differentiate between normal pancreatic tissue and PDAC. This would help in detecting small PDACs (<2 cm) early on CT scans, directing timely interventions. Chu et al,40 Chen et al,41 and Wang et al46 have demonstrated promising results in differentiating PDAC from normal pancreatic tissue using radiomic features. Chu et al analyzed manually segmented CT images and trained a random forest binary classification model A model to differentiate between PDAC and normal tissue. They found that a radiomics-based classification model achieved 100% sensitivity in the detection of PDAC. However, their study was limited by small sample size, variations in scanning protocols, and age mismatch between healthy controls and PDAC patients, which may have influenced radiomics-based classification due to age-related fatty and atrophic pancreatic changes.
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579. - Abdominal CT is the most frequently used imaging technique for different intraabdominal conditions. This results in large datasets and allows for a retrospective examination of “normal pre-diagnostic CT scans” in patients who subsequently developed a PDAC. Pre-diagnostic CT scans are defined as incidental CT scans performed for unrelated indications before PDAC diagnosis. Javed et al trained a naïve Bayes classification model employing radiomics to differentiate between high-risk and low-risk pancreas on pre-diagnostic CT scans. They achieved a classification accuracy of 93% on the internal dataset and concluded that early local textural changes, which can be detected by radiomics, occurred in the sub-region of the pancreas corresponding to tumor development later.
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579. - Mukherjee et al further validated the effectiveness of radiomics-based ML models and demonstrated that AI can detect PDAC at a median of 386 (range: 97-1092) days before clinical diagnosis by identifying subtle changes that are imperceptible to the human eye. They extracted 88 first order and gray-level texture features and evaluated 4 ML classifiers, with SVM emerging as the best performer with an AUC of 0.98 (95% CI: 0.94, 0.98) and accuracy of 92% (95% CI: 86.7, 93.7). Their AI models significantly outperformed radiologists, who exhibited only fair inter-reader agreement (Cohen’s kappa, 0.3) and a substantially lower diagnostic performance with an AUC of 0.66 (95% CI: 0.46, 0.86).
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579. - “While radiomics possesses significant potential, there are several drawbacks that hinder its practical application. Developing a radiomic pipeline necessitates the processing of large imaging datasets, which in turn demands substantial computational resources, efficient data storage and management systems, and skilled personnel. A dedicated multidisciplinary team, comprising radiologists, data scientists, and engineers, is essential for the development and validation of the radiomic workflow. This extensive workflow is both time-consuming and resource intensive, heavily dependent on the expertise of the production team. There is a need for user-independent programs that clinicians can operate with ease.”
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579. - Many radiomics studies do not justify their sample size, leading to models that may be too small to avoid overfitting, limiting their widespread adoption and reproducibility. These models may yield very high accuracy on training datasets and perform poorly on external validation sets. There is a need for large, multi-center studies to bridge this validation gap, however, data sharing among institutions remains a critical barrier. Ahmed et al assessed the radiomics quality score across 54 studies focused on pancreatic imaging and reported a low mean score of 32.1%, reflecting significant methodological heterogeneity and limited reproducibility of the current Radiomics models. Notably, none of the studies included a cost-effectiveness analysis, raising concerns about clinical integration of these models. Additionally, radiomics-based Studies have shown low adherence to reporting guidelines and quality frameworks, including IBSI and CLAIM (Checklist for Artificial Intelligence in Medical Imaging), further limiting The study's reproducibility.
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579.- The challenges are multifaceted, stemming from the complexities involved in constructing and validating a radiomics model. The radiomics pipeline is highly sensitive to data variations introduced by differing scan protocols and scanner manufacturers, limiting the construction of a robust reproducible radiomics model. Furthermore, leading vendors such as Siemens Healthineers, GE Healthcare, Phillips, Canon, and United Imaging offer a range of 5 to 10 different imaging systems, each utilizing distinct algorithms, which adds to the complexity of the issue. Even with the same vendors, institutions tend to have institution-specific protocols that contribute to additional variability. To address these challenges, future efforts could focus on the development of parallel image reconstruction pipelines, one already existing optimized for human evaluation and another specifically tailored to support AI model development.
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579. - Radiomics has revolutionized the paradigm of early detection of PDAC with promising results but remains largely unused in clinical practice. Bridging the gap to clinical implementation will require overcoming challenges in both the radiomics workflow and the clinical side. Standardized imaging protocols, optimized image filters, and reconstruction parameters tailored for radiomics could help develop robust and reproducible models. However, the binary output of the model, which is dependent on the training set, restricts its ability to detect different pancreatic lesions, thus limiting its real-world applicability. Due to the overall low incidence of PDAC, routine screening is not recommended, but radiomics combined with multi-cancer early detection may assist in early detection using imaging data obtained for other causes. While this is an interesting concept to explore, the generation of a high number of false positives remains a significant concern.
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579.
Radiomics in Early Detection of Pancreatic Ductal Adenocarcinoma: A Close Look at Its Current Status and Challenges to Clinical Implementation.
Arshad H, Lopez-Ramirez F, Tixier F, Soyer P, Kawamoto S, Fishman EK, Chu LC
Can Assoc Radiol J. 2025 Jul Epub ahead of print. PMID: 40632579.
- Purpose To determine whether renal cell carcinoma metastases (RCC-Mets) to the pancreas can be differentiated from pancreatic neuroendocrine tumors (PNETs) in patients with RCC on CT or MRI at presentation
Conclusion PNETs are typically single, occur in distal pancreas, and enhance homogeneously compared to RCC-Mets which are often multiple, occur in the proximal pancreas, and enhance heterogeneously.
Differentiation between renal cell carcinoma metastases to the pancreas and pancreatic neuroendocrine tumors in patients with renal cell carcinoma on CT or MRI.
Nduwimana MJ, Colak C, Bilgin C, et al.
Abdom Radiol (NY). 2025 Jul;50(7):3115-3125.. - Results The study group comprised of 134 patients (90 males) with 265 lesions (229 RCC-Mets and 36 PNETs). Patients with PNETs were significantly younger (62 ± 12 years vs. 67 ± 9 years, p = 0.013). Inter-observer agreement for CT/MRI features was excellent across multiple imaging variables (k = 0.86–1.00). Most PNETs were single lesions (88 vs. 63%, p = 0.008), smaller in size (14 mm vs. 23 mm, p = 0.042), more common in the body and tail (81 vs. 57%, p = 0.01), showed homogeneous contrast enhancement (64–79% vs. 39–49%, p < 0.01–0.03), less T1-hypointense (80 vs. 99%, p = 0.002) and more DWI hyperintense (71 vs. 58%, p < 0.001) compared to RCC-Mets.
Differentiation between renal cell carcinoma metastases to the pancreas and pancreatic neuroendocrine tumors in patients with renal cell carcinoma on CT or MRI.
Nduwimana MJ, Colak C, Bilgin C, et al.
Abdom Radiol (NY). 2025 Jul;50(7):3115-3125.. - Metastases to the pancreas are rare with an estimated incidence of 2–5% in clinical series . Common malignancies that metastasize to the pancreas include renal cell carcinoma (RCC), lung carcinomas, breast carcinoma, colorectal carcinoma, melanoma and soft-tissue sarcomas . RCC is the most common primary neoplasm to metastasize to the pancreas, accounting for approximately 30% of pancreatic metastases . RCC is also the most common malignancy to present with isolated pancreatic metastases. Though rare, pancreatic neuroendocrine tumors (PNETs) are the second most common primary pancreatic neoplasm with varying biologic behavior from benign to malignant.
Differentiation between renal cell carcinoma metastases to the pancreas and pancreatic neuroendocrine tumors in patients with renal cell carcinoma on CT or MRI.
Nduwimana MJ, Colak C, Bilgin C, et al.
Abdom Radiol (NY). 2025 Jul;50(7):3115-3125.. - PNETs more commonly developed in younger patients (62 ± 12 years vs. 67 ± 9 years, p = 0.013), in males [27/32 (84.4%) vs. 63/102 (61.8%), p = 0.017], and were diagnosed earlier following nephrectomy (5.3 ± 9.7 years vs. 9.9 ± 7.4 years, p = 0.007) compared to the RCC-Mets group (Table 1). The number of pancreatic lesions ranged from 1 to 27, with a mean ± SD of 2 ± 2.9 per patient. There was no significant difference in the number of lesions per patient between RCC-Mets and PNETs (2.2 ± 3.2 vs. 1.2 ± 0.4, p = 0.062). However, PNETs more commonly presented as single lesions (28/32 = 88% vs. 64/102 = 63%, p = 0.008), and were more frequently located in the distal pancreas (80.6 vs. 57.2%, p = 0.01), compared to RCC-Mets,
Differentiation between renal cell carcinoma metastases to the pancreas and pancreatic neuroendocrine tumors in patients with renal cell carcinoma on CT or MRI.
Nduwimana MJ, Colak C, Bilgin C, et al.
Abdom Radiol (NY). 2025 Jul;50(7):3115-3125.. - In our study, though most of the PNETs were histopathologically proven grade 1, nearly half of the cases underwent,resection. However, current management for PNETs suggests that surgery be driven by tumor size (> 2 cm) and/ or poor histopathologic differentiation (grade 3) as the disease is mostly benign. In contrast, most RCC-Mets underwent pathologic confirmation through biopsy and only 54% of patients underwent surgical resection, despite it being the primary treatment for the disease. Resection is linked to favorable prognosis and extended survival . Moreover, pancreatic biopsy is a challenging and invasive procedure due to the pancreas’ anatomic location. The above findings may be attributed to the two-decade span of our study, during which treatment guidelines have evolved. Accurately distinguishing RCC-Mets from PNETs on imaging could facilitate earlier surgical intervention and potentially improving survival outcomes. Additionally, with RCCMets presenting even up to two decades after nephrectomy, routine, long-term, radiologic surveillance is encouraged.
Differentiation between renal cell carcinoma metastases to the pancreas and pancreatic neuroendocrine tumors in patients with renal cell carcinoma on CT or MRI.
Nduwimana MJ, Colak C, Bilgin C, et al.
Abdom Radiol (NY). 2025 Jul;50(7):3115-3125.. - Lastly, several different hypervascular lesions can occur within and around pancreas that need to be differentiated from RCC-Mets and PNETs, including intrapancreatic accessory spleen, solid pseudopapillary tumor, solid serous cystadenoma, acinar cell carcinoma, hypervascular metastases, and gastrointestinal stromal tumor (GIST). Intrapancreatic accessory spleens follow the enhancement pattern of the spleen on all contrast enhanced phases and have similar signal intensity as spleen parenchyma on MRI. Solid pseudopapillary tumors almost exclusively occurs in young females and may present with heterogeneous enhancement and T1 hyperintensity due to hemorrhage. Solid appearing serous cystadenomas typically occur in asymptomatic elderly woman and may rarely appear solid; however, characteristic small cystic spaces may be visible on MRI differentiating from other lesions.
Differentiation between renal cell carcinoma metastases to the pancreas and pancreatic neuroendocrine tumors in patients with renal cell carcinoma on CT or MRI.
Nduwimana MJ, Colak C, Bilgin C, et al.
Abdom Radiol (NY). 2025 Jul;50(7):3115-3125.. - In conclusion, there are significant differences between RCC-Mets and PNET lesions in patients with RCC on CT or MRI. PNETs are usually single, seen in the body and tail, and exhibit homogeneous enhancement. Whereas RCC-Mets to the pancreas are more often multiple, seen in the head, uncinate and neck region of the pancreas, and exhibit heterogeneous enhancement. These imaging differences may be useful for early diagnosis to triage appropriate follow-up management.
Differentiation between renal cell carcinoma metastases to the pancreas and pancreatic neuroendocrine tumors in patients with renal cell carcinoma on CT or MRI.
Nduwimana MJ, Colak C, Bilgin C, et al.
Abdom Radiol (NY). 2025 Jul;50(7):3115-3125..
Small Bowel
- IgG4-related disease (IgG4-RD) is a chronic inflammatory condition characterized by tissue infiltration with IgG4-positive plasma cells, leading to fibrosis and organ dysfunction. While primarily affecting the pancreas, bile ducts, and salivary glands, IgG4-RD can also involve the gastrointestinal tract, raising questions about its relationship with inflammatory bowel disease (IBD). Recent studies suggest that patients with IBD may exhibit histological and serological features consistent with IgG4-RD, such as a dense lymphoplasmacytic infiltration, a storiform-type of fibrosis and a prominent IgG4 immune response. This overlap represents a diagnostic challenge, as differentiating between primary IBD and IgG4-RD involving the gut is crucial for appropriate treatment.
IgG4 in the gut: Gastrointestinal IgG4-related disease or a new subtype of inflammatory bowel disease.
Bencardino S, et al.
Autoimmun Rev. 2025 Jan 31;24(2):103720. - • Intestinal involvement in IgG4-related disease (IgG4-RD) can resemble conditions like Crohn's disease, causing diagnostic difficulties due to overlapping features
•Elevated IgG4+ plasma cells in inflammatory bowel disease (IBD) may be associated with a more aggressive subtype, marked by increased inflammation, greater risk of complications, and possibly requiring more intensive management.
•Understanding the role of serum and tissue IgG4 in IBD is crucial for implementing personalized medicine approaches.
IgG4 in the gut: Gastrointestinal IgG4-related disease or a new subtype of inflammatory bowel disease.
Bencardino S, et al.
Autoimmun Rev. 2025 Jan 31;24(2):103720. - Intestinal manifestations of IgG4-RD can mimic inflammatory bowel disease (IBD) . IBD, comprising Crohn's Disease (CD) and ulcerative colitis (UC), is a group of chronic, relapsing conditions resulting in inflammation, and in some cases fibrosis, of the gastrointestinal tract . Although the prevalence of IBD has traditionally been higher in developed countries, such as Europe and North America, their incidence and prevalence is rapidly increasing also in developing countries, indicating its emergence as a global health concern .
IgG4 in the gut: Gastrointestinal IgG4-related disease or a new subtype of inflammatory bowel disease.
Bencardino S, et al.
Autoimmun Rev. 2025 Jan 31;24(2):103720. - In conclusion, IBD and IgG4-RD are fibro-inflammatory diseases requiring an appropriate diagnostic workup as well as long-term monitoring. Serum and tissue IgG4 assessment could be promising in diagnosing and monitoring these conditions. However, large prospective studies are needed to confirm the role of IgG4 in IBD, and to support their use in clinical practice. A multidisciplinary approach involving the collaboration of gastroenterologists, pathologists, and immunologists who are specialized in IgG4-RD is highly recommended to ensure precise diagnosis and tailored treatment strategies.
IgG4 in the gut: Gastrointestinal IgG4-related disease or a new subtype of inflammatory bowel disease.
Bencardino S, et al.
Autoimmun Rev. 2025 Jan 31;24(2):103720. 
IgG4-related diseases of the digestive tract.
Löhr JM, Vujasinovic M, Rosendahl J, Stone JH, Beuers U.
Nat Rev Gastroenterol Hepatol. 2022 Mar;19(3):185-197.
Spleen
- “Practice Recommendation: Splenic artery aneurysms ≥2 cm, or any aneurysm with features suspicious for a pseudoaneurysm should be referred to interventional radiology (or other endovascular specialist) for consideration of treatment. Aneurysms <2 cm can be followed for growth annually with CT or MR angiography, with discontinuation of follow-up made in consultation with a vascular specialist after a period of ongoing stability.”
CAR Recommendations for the Management of Incidental Findings of the Spleen and Nodes in Adults
Jeffery R. Bird, Gary L. Brahm, Christopher I. Fung et al.
Canadian Association of Radiologists Journal1–10 (2025) - “Practice Recommendation: A single measurement of >13 cm in maximal diameter is recommended to screen for splenomegaly in adults, recognizing that the positive predictive value for disease has not been determined. Volume calculations can be reserved for when more accuracy is required.”
CAR Recommendations for the Management of Incidental Findings of the Spleen and Nodes in Adults
Jeffery R. Bird, Gary L. Brahm, Christopher I. Fung et al.
Canadian Association of Radiologists Journal1–10 (2025) - Although medical calculators can diagnose splenomegaly by correcting for body size, they are cumbersome, requiring knowledge of the patient’s height, weight, and gender.6 Although the literature suggests that splenic volume calculation may represent the future of spleen measurement, other studies showing a close correlation between a single largest measurement and total spleen volume favour continuing with the current status quo of providing a single value to represent spleen size. Volume calculations can be referenced to body size when more accuracy is required, particularly to avoid overdiagnosing splenomegaly in larger patients.
CAR Recommendations for the Management of Incidental Findings of the Spleen and Nodes in Adults
Jeffery R. Bird, Gary L. Brahm, Christopher I. Fung et al.
Canadian Association of Radiologists Journal1–10 (2025) - “Lymphoma is the most common malignancy of the spleen, either primary or part of diffuse systemic disease.12,15 Splenic involvement occurs in approximately 33% of patients with Hodgkins and 30% to 40% in patients with non-Hodgkins Lymphoma.16 Lymphoma can present in many forms including splenomegaly, diffuse nodules (either in a miliary pattern or larger nodules), or a solitary mass.12 Primary splenic lymphoma confined only to the spleen ± perisplenic nodes is very rare, comprising less than 1% of cases, and most patients will present with constitutional symptoms.”
CAR Recommendations for the Management of Incidental Findings of the Spleen and Nodes in Adults
Jeffery R. Bird, Gary L. Brahm, Christopher I. Fung et al.
Canadian Association of Radiologists Journal1–10 (2025) - Splenic incidental findings are defined as lesions detected on imaging in the spleen not related to the clinical history. Incidental splenic lesions are less common than in other organs such as liver or kidneys, but increased demand for imaging means that their frequency is rising. Incidental focal splenic lesions have a wide range of etiologies, ranging from common benign diagnoses (cysts, granulomas, and hemangiomas) to lymphoma or metastases to exceedingly rare primary malignancies such as angiosarcoma. In one study, 1.5% of trauma patients with CT had an incidental splenic lesion and the vast majority are benign. Benign lesions are almost always asymptomatic, whereas malignant lesions are very rarely entirely incidental or a solitary isolated finding.
CAR Recommendations for the Management of Incidental Findings of the Spleen and Nodes in Adults
Jeffery R. Bird, Gary L. Brahm, Christopher I. Fung et al.
Canadian Association of Radiologists Journal1–10 (2025) - Practice Recommendation: If an incidental isolated indeterminate splenic mass is found on CT or MR in a patient with no history of malignancy or symptoms, it is unlikely to be clinically significant, and no further evaluation or follow-up is necessary.
Practice Recommendation: In patients with constitutional symptoms (fever, weight loss, night sweats), epigastric or left upper quadrant pain, or a history of prior malignancy, the risk of malignancy is low but not negligible. An incidental indeterminate splenic lesion should be further evaluated with MRI, PET/CT, or biopsy, especially if it may affect patient management.
CAR Recommendations for the Management of Incidental Findings of the Spleen and Nodes in Adults
Jeffery R. Bird, Gary L. Brahm, Christopher I. Fung et al.
Canadian Association of Radiologists Journal1–10 (2025)
Vascular
- “Practice Recommendation: Splenic artery aneurysms ≥2 cm, or any aneurysm with features suspicious for a pseudoaneurysm should be referred to interventional radiology (or other endovascular specialist) for consideration of treatment. Aneurysms <2 cm can be followed for growth annually with CT or MR angiography, with discontinuation of follow-up made in consultation with a vascular specialist after a period of ongoing stability.”
CAR Recommendations for the Management of Incidental Findings of the Spleen and Nodes in Adults
Jeffery R. Bird, Gary L. Brahm, Christopher I. Fung et al.
Canadian Association of Radiologists Journal1–10 (2025) - “Practice Recommendation: A single measurement of >13 cm in maximal diameter is recommended to screen for splenomegaly in adults, recognizing that the positive predictive value for disease has not been determined. Volume calculations can be reserved for when more accuracy is required.”
CAR Recommendations for the Management of Incidental Findings of the Spleen and Nodes in Adults
Jeffery R. Bird, Gary L. Brahm, Christopher I. Fung et al.
Canadian Association of Radiologists Journal1–10 (2025)