Imaging Pearls ❯ Liver ❯ Artificial Intelligence
|
-- OR -- |
|
- “Diagnosing liver lesions is crucial for treatment choices and patient outcomes. This study develops an automatic diagnosis system for liver lesions using multiphase enhanced computed tomography (CT). A total of 4039 patients from six data centers are enrolled to develop Liver Lesion Network (LiLNet). LiLNet identifies focal liver lesions, including hepatocellular carcinoma (HCC), intrahepatic cholangiocarcinoma (ICC), metastatic tumors (MET), focal nodular hyperplasia (FNH), hemangioma (HEM), and cysts (CYST). Validated in four external centers and clinically verified in two hospitals, LiLNet achieves an accuracy (ACC) of 94.7% and an area under the curve (AUC) of 97.2% for benign and malignant tumors. For HCC, ICC, and MET, the ACC is 88.7% with an AUC of 95.6%. For FNH, HEM, and CYST, the ACC is 88.6% with an AUC of 95.9%. LiLNet can aid in clinical diagnosis, especially in regions with a shortage of radiologists.”
Focal liver lesion diagnosis with deep learning and multistage CT imaging
Yi Wei et al.
Nature Communications | ( 2024) 15:7040 - “In this study, we demonstrate the performance of our AI system, LiLNet, in distinguishing six common types of focal liver lesions. We develop the model using data from six centers and assess its generalization through extensive testing on a test set and four externalvalidation centers. We compare LiLNet’s performance with radiologists’ interpretations of contrast-enhanced CT images in a reader study. To address real-world clinical implementation, we deploy LiLNet in two hospitals, integrating it into routine workflows across outpatient, emergency, and inpatient settings. This integration evaluates the system’s performance in various clinical environments, ensuring its robustness and reliability in practical use.”
Focal liver lesion diagnosis with deep learning and multistage CT imaging
Yi Wei et al.
Nature Communications | ( 2024) 15:7040 - “We used a test set of 6743 images from 221 patients at West China Hospital of Sichuan University to compare the diagnostic ability of LiLNet with that of radiologists. The evaluation involved three radiologists with varying levels of experience. Radiologists independently labeled the 221 patients based on multiphase contrast-enhanced CT images. LiLNet demonstrated a diagnostic accuracy of 91.0% for distinguishing between benign and malignant tumors, 82.9% for distinguishing between malignant tumors, and 92.3% for distinguishing between benign tumors. Compared to junior-level radiologists, LiLNet achieved 4.6% greater accuracy for benign and malignant diagnosis, 4.1% greater accuracy for middle-level radiologists, and 2.3% greater accuracy for senior level radiologists. The diagnostic accuracy of radiologists for diagnosing malignant tumors was similar. Notably, compared with radiologists, LiLNet achieved a substantial 18% improvement in diagnostic accuracy. Additionally, in diagnosing benign tumors, LiLNet outperformed junior-level practitioners by 20%, middle-level practitioners by 10%, and senior-level practitioners by 6.7%.”
Focal liver lesion diagnosis with deep learning and multistage CT imaging
Yi Wei et al.
Nature Communications | ( 2024) 15:7040 
Focal liver lesion diagnosis with deep learning and multistage CT imaging
Yi Wei et al.
Nature Communications | ( 2024) 15:7040- “We utilized patients’ name-ID as the unique identifier to prevent duplicate IDs. Duplicate samples with the same name-ID were systematically removed, and patients were randomly assigned to either the training set or testing set to prevent data overlap. As shown in Fig. 1a, the training set comprised images from 1580 patients from West ChinaHospital of SichuanUniversity and Sanya People’sHospital. The testing cohort consisted of 1308 patients from West China Hospital of Sichuan University, while external validation cohorts included 1151 patients from Henan Provincial People’s Hospital, The First Affiliated Hospital of Chengdu Medical College, Leshan People’s Hospital, and Guizhou Provincial People’s Hospital.”
Focal liver lesion diagnosis with deep learning and multistage CT imaging
Yi Wei et al.
Nature Communications | ( 2024) 15:7040 - “Between June 2012 and December 2022, a total of 4039 patients’ multiphase (arterial phase and portal venous phase) contrastenhanced CT images from six hospitals were included under the following inclusion criteria: Patients (1) were eighteen years or older; (2) did not have a history of hepatectomy, transarterial chemotherapy (TACE), or radiofrequency ablation (RFA) before CT imaging; (3) had pathologically confirmed malignant tumors; and (4) had benign tumors confirmed either by consensus among three radiologists or by follow-up of at least six months using two imaging modalities. The method used for retrospective data collection and basic patientinformation including sex and age are depicted in Fig. 1a and Table 1, respectively. Furthermore, clinical testing was conducted on two real world clinical evaluation queues (Fig. 1b): West China Tianfu Center and Sanya People’s Hospital. At Tianfu Center, we examined 184 cases, while at Sanya People’s Hospital, 235 cases were assessed. Gender andAge assignment was based on government-issued IDs.”
Focal liver lesion diagnosis with deep learning and multistage CT imaging
Yi Wei et al.
Nature Communications | ( 2024) 15:7040
- Purpose: To evaluate the sensitivity of artificial intelligence (AI)-powered software in detecting liver metastases, especially those overlooked by radiologists.
Results: The software successfully processed images from 135 patients. The per-lesion sensitivity for all liver lesion types, liver metastases, and liver metastases overlooked by radiologists was 70.1%, 70.8%, and 55.0%, respectively. The software detected liver metastases in 92.7% and 53.7% of patients in detected and overlooked cases, respectively. The average number of false positives was 0.48 per patient.
Conclusion: The AI-powered software detected more than half of liver metastases overlooked by radiologists while maintaining a relatively low number of false positives. Our results suggest the potential of AI-powered software in reducing the frequency of overlooked liver metastases when used in conjunction with the radiologists’ clinical interpretation.
Artificial intelligence-powered software detected more than half of the liver metastases overlooked by radiologists on contrast-enhanced CT
Hirotsugu Nakai et al.
European Journal of Radiology 163 (2023) 110823 - “The AI-powered software detected more than half of liver metastases overlooked by radiologists while maintaining a relatively low number of false positives. Our results suggest the potential of AI-powered software in reducing the frequency of overlooked liver metastases when used in conjunction with the radiologists’ clinical interpretation. ”
Artificial intelligence-powered software detected more than half of the liver metastases overlooked by radiologists on contrast-enhanced CT
Hirotsugu Nakai et al.
European Journal of Radiology 163 (2023) 110823 - The per-lesion sensitivity for liver lesions was as follows: all liver lesion types, 70.1% (700/999); liver metastases, 70.8% (155/ 219); liver metastases detected by radiologists, 84.0% (100/119); and liver metastases overlooked by radiologists, 55.0% (55/100). Figs. 2–4 present examples of liver metastases correctly detected only by the software, correctly detected only by radiologists, and overlooked by both the software and radiologists. A total of 65 false positives were identified across 48 patients, with 27.7% (18/65) of false positives in the diaphragm, 24.6% (16/65) in the focal fatty liver, and 15.3% (10/65) in the bile ducts.
Artificial intelligence-powered software detected more than half of the liver metastases overlooked by radiologists on contrast-enhanced CT
Hirotsugu Nakai et al.
European Journal of Radiology 163 (2023) 110823 - “The results have indicated its potential in reducing the number of overlooked liver metastases and providing reference information to the investigating radiologists. Meanwhile, the software did not detect 16% of liver metastases found by radiologists. For instance, the software failed to detect liver metastases in contact with large hepatic veins, whereas the radiologists did not; this leads us to believe that lesions surrounded by the liver parenchyma may be easier for the software to detect than those that are not. Therefore, the software would not replace the radiologists; instead, the software should be used in conjunction with the radiologists’ clinical interpretation. Nonetheless, this is only the first version of the software; the next version is expected to show improvements in detection performance and new functionalities. Given that the number of false positives was relatively low (0.48 per patient), lowering the threshold for liver lesions should be considered from the viewpoint of screening to identify lesion candidates. Additionally, radiologists can also make false-positive diagnoses. ”
Artificial intelligence-powered software detected more than half of the liver metastases overlooked by radiologists on contrast-enhanced CT
Hirotsugu Nakai et al.
European Journal of Radiology 163 (2023) 110823
- IMPORTANCE In patients with resectable colorectal cancer liver metastases (CRLM), the choice of surgical technique and resection margin are the only variables that are under the surgeon’s direct control and may influence oncologic outcomes. There is currently no consensus on the optimal margin width.
OBJECTIVE To determine the optimal margin width in CRLM by using artificial intelligence–based techniques developed by the Massachusetts Institute of Technology and to assess whether optimal margin width should be individualized based on patient characteristics.
CONCLUSIONS AND RELEVANCE This cohort study used artificial intelligence–based methodologies to provide a possible resolution to the long-standing debate on optimal margin width in CRLM.
Using Artificial Intelligence to Find the Optimal Margin Width in Hepatectomy for Colorectal Cancer Liver Metastases
Dimitris Bertsimas et al.
JAMA Surg. doi:10.1001/jamasurg.2022.1819 - OBJECTIVE To determine the optimal margin width in CRLM by using artificial intelligence–based techniques developed by the Massachusetts Institute of Technology and to assess whether optimal margin width should be individualized based on patient characteristics.
DESIGN, SETTING, AND PARTICIPANTS The internal cohort of the study included patients who underwent curative-intent surgery for KRAS-variant CRLM between January 1, 2000, and December 31, 2017, at Johns Hopkins Hospital, Baltimore, Maryland, Memorial Sloan Kettering Cancer Center, New York, New York, and Charité–University of Berlin, Berlin,Germany. Patients from institutions in France, Norway, the US, Austria, Argentina, and Japan were retrospectively identified from institutional databases and formed the external cohort of the study. Data were analyzed from April 15, 2019, to November 11, 2021.
Using Artificial Intelligence to Find the Optimal Margin Width in Hepatectomy for Colorectal Cancer Liver Metastases
Dimitris Bertsimas et al.
JAMA Surg. doi:10.1001/jamasurg.2022.1819 - RESULTS This cohort study included a total of 1843 patients (internal cohort, 965; external cohort, 878). The internal cohort included 386 patients (median [IQR] age, 58.3 [49.0-68.7]years; 200 men [51.8%]) with KRAS-variant tumors. The AUC of the RF counterfactual model was 0.76 in both the internal training and testing cohorts, which is the highest ever reported. The recommended optimal margin widths for patient subgroups A, B, C, and D were 6, 7, 12, and 7 mm, respectively. The SHAP analysis largely confirmed this by suggesting 6 to 7mmforsubgroup A, 7mmfor subgroup B, 7 to 8mmfor subgroup C, and 7mmf or subgroup D. The external cohort included 375 patients (median [IQR] age, 61.0 [53.0-70.0] years; 218 men [58.1%]) with KRAS-variant tumors. The new RF model had an AUC of 0.78, which allowed fora reliable external validation of the OPT-based optimal margin. The external validation was successful as it confirmed the association of the optimal margin width of 7mmwith a considerable prolongation of survival in the external cohort.
CONCLUSIONS AND RELEVANCE This cohort study used artificial intelligence–based methodologies to provide a possible resolution to the long-standing debate on optimal margin width in CRLM.
Using Artificial Intelligence to Find the Optimal Margin Width in Hepatectomy for Colorectal Cancer Liver Metastases
Dimitris Bertsimas et al.
JAMA Surg. doi:10.1001/jamasurg.2022.1819
- “Cinematic rendering (CR) is a recently described three-dimensional (3D) rendering technique that generates photorealistic images based on a new lighting model. This review illustrates the potential application of CR in the evaluation of focal liver masses. CR shows promise in improving the visualization of enhancement pattern and internal architecture, local tumor extension, and global disease burden, which may be helpful in focal liver mass characterization and pretreatment planning.”
Cinematic rendering of focal liver masses
L.C. Chu, S.P. Rowe, E.K. Fishman
Diagnostic and Interventional Imaging,Volume 100, Issue 9,2019,Pages 467-476
- “Artificial intelligence is poised to revolutionize medical image. It takes advantage of the high-dimensional quantitative features present in medical images that may not be fully appreciated by humans. Artificial intelligence has the potential to facilitate automatic organ segmentation, disease detection and characterization, and prediction of disease recurrence. This article reviews the current status of artificial intelligence in liver imaging and reviews the opportunities and challenges in clinical implementation.”
Current Status of Radiomics and Deep Learning in Liver Imaging
Linda C. Chu, Seyoun Park, Satomi Kawamoto, Alan L. Yuille, Ralph H. Hruban, and Elliot K. Fishman
J Comput Assist Tomogr 2021;45: 343–351 - "Artificial intelligence is commonly performed with supervised learning because of the complexity of medical image analysis. It is provided with annotated “ground truth,” which is used as feedback to improve the algorithm. The degree of data annotation in a detection or segmentation problem can range from labeling subjects as normal versus abnormal, creating an approximate bounding box in the region of the abnormality, to detailed slice-by-slice segmentation of the specific abnormality. Artificial intelligence algorithms depend on large data sets with high-quality images and annotations for training as well as validation, and their performance increases logarithmically with increased training data.”
Current Status of Radiomics and Deep Learning in Liver Imaging
Linda C. Chu, Seyoun Park, Satomi Kawamoto, Alan L. Yuille, Ralph H. Hruban, and Elliot K. Fishman
J Comput Assist Tomogr 2021;45: 343–351 - “Radiomics converts imaging data into high-dimensional quanitative features, which can be classified into first order, shape, and texture features. First-order features are derived from histogram distribution of individual voxel signal intensities, which can provide statistics on central tendency, variance, range, and shape of the distribution. Shape features are generated from the 3-dimensional surface mask of the region of interest and can provide measures such as volume, surface area, and sphericity. Texture features, also referred to as second order features, quantify the correlation of signal intensities with respect to surrounding voxels in 3 dimensions. In addition, different types of filters (eg, wavelets, Laplacian of Gaussian) are often applied to the original imaging volume to generate the filtered imaging volume before feature extraction. This process typically generates hundreds of features, and redundant features are eliminated through dimension reduction. Machine learning algorithms, such as random forest and support vector machine, are frequently used to analyze the most relevant features.”
Current Status of Radiomics and Deep Learning in Liver Imaging
Linda C. Chu, Seyoun Park, Satomi Kawamoto, Alan L. Yuille, Ralph H. Hruban, and Elliot K. Fishman
J Comput Assist Tomogr 2021;45: 343–351 - “Deep learning-based algorithms have outperformed traditional semiautomatic interactive methods25 and can be significantly faster than manual segmentation.26 In most cases, the deep network is trained to segment the liver in a specific modality (eg, CT or MRI). Wang et al16 trained an algorithm on 1 modality (eg, unenhanced MRI) and was able to adapt it to other modalities (eg, contrast-enhanced CT or MRI) via transfer learning. This type of modality-independent segmentation algorithm may broaden the scope of potential clinical applications.”
Current Status of Radiomics and Deep Learning in Liver Imaging
Linda C. Chu, Seyoun Park, Satomi Kawamoto, Alan L. Yuille, Ralph H. Hruban, and Elliot K. Fishman
J Comput Assist Tomogr 2021;45: 343–351 - "Because of black-box nature of the AI models, it can be difficult for the radiologists and clinicians to understand the rationale behind the AI output. This can especially problematic if there is discrepancy between the radiologists' subjective assessment and the AI output. Wang et al sought to make deep learning classification results more “understandable” by training a convolutional neural network with specific training image examples of radiologic features relevant in liver lesion classification. They generated feature maps that ranked the most relevant features used in the lesion classification task. This type of supportive evidence may increase radiologists' confidence in the AI classification.”
Current Status of Radiomics and Deep Learning in Liver Imaging
Linda C. Chu, Seyoun Park, Satomi Kawamoto, Alan L. Yuille, Ralph H. Hruban, and Elliot K. Fishman
J Comput Assist Tomogr 2021;45: 343–351 - "Radiomics models have been used to predict recurrence and survival after curative resection, ablation or liver transplantatio8 in patients with HCC. Radiomics models have been able to predict disease recurrence with higher accuracy than traditional clinical models, with the C-index (concordance index measuring goodness of fit for binary outcomes) ranging from 0.47 to 0.82 for radiomics models versus 0.56 to 0.78 for clinical models. The addition of radiomics features to clinical models generally helps to improve risk stratification. It makes intuitive sense that the best prognostic models would incorporate both tumor features (captured by qualitative imaging features and radiomics features) and clinical features that consider background liver disease, patient comorbidities, and performance status.”
Current Status of Radiomics and Deep Learning in Liver Imaging
Linda C. Chu, Seyoun Park, Satomi Kawamoto, Alan L. Yuille, Ralph H. Hruban, and Elliot K. Fishman
J Comput Assist Tomogr 2021;45: 343–351 - "Legal ramifications also need to be considered in the clinical implementation of AI. Just like humans, AI is not perfect. The IBM Watson Health's cancer AI algorithm (known as Watson for Oncology) was trained on a small number of synthetic cases with limited input from oncologists. As a result, many output treatment recommendations were erroneous and potentially harmful.93 If a clinician makes a mistake based on flawed AI decision support, who is legally responsible? The black-box nature of many AI algorithms also complicates efforts to tease out the exact cause of any errors. Medical AI systems are too new to have been involved in malpractice lawsuits, and it remains to be seen where the responsibilities lie.”
Current Status of Radiomics and Deep Learning in Liver Imaging
Linda C. Chu, Seyoun Park, Satomi Kawamoto, Alan L. Yuille, Ralph H. Hruban, and Elliot K. Fishman
J Comput Assist Tomogr 2021;45: 343–351 - "In this early exploratory phase of applying AI to liver imaging, studies have shown that AI can achieve impressive performance in disease detection, classification, and prognostication under highly controlled experimental settings. These results should be validated in multicenter trials with stringent postmarketing monitoring to ensure safety and efficacy across different practice environments. Artificial intelligence algorithms should be combined to perform comprehensive organ-specific or more general abdominal abnormality detection to be more clinically relevant.”
Current Status of Radiomics and Deep Learning in Liver Imaging
Linda C. Chu, Seyoun Park, Satomi Kawamoto, Alan L. Yuille, Ralph H. Hruban, and Elliot K. Fishman
J Comput Assist Tomogr 2021;45: 343–351 
Current Status of Radiomics and Deep Learning in Liver Imaging
Linda C. Chu, Seyoun Park, Satomi Kawamoto, Alan L. Yuille, Ralph H. Hruban, and Elliot K. Fishman
J Comput Assist Tomogr 2021;45: 343–351
Current Status of Radiomics and Deep Learning in Liver Imaging
Linda C. Chu, Seyoun Park, Satomi Kawamoto, Alan L. Yuille, Ralph H. Hruban, and Elliot K. Fishman
J Comput Assist Tomogr 2021;45: 343–351
Current Status of Radiomics and Deep Learning in Liver Imaging
Linda C. Chu, Seyoun Park, Satomi Kawamoto, Alan L. Yuille, Ralph H. Hruban, and Elliot K. Fishman
J Comput Assist Tomogr 2021;45: 343–351
- OBJECTIVE To examine whether deep learning recurrent neural network(RNN )models that use raw longitudinal data extracted directly from electronic health records outperform conventional regression models in predicting the risk of developing hepatocellular carcinoma (HCC).
CONCLUSIONS AND RELEVANCE In this study, deep learning RNN models outperformed conventional LR models, suggesting that RNN models could be used to identify patients with HCV-related cirrhosis with a high risk of developing HCC for risk-based HCC outreach and surveillance strategies.
Assessment of a Deep Learning Model to Predict Hepatocellular Carcinoma in Patients With Hepatitis C Cirrhosis
George N. Ioannou et al.
JAMA Network Open. 2020;3(9):e2015626. Sept 2020 - “This study has limitations related to lack of external validation and the computational cost of running the analyses. To reduce computational cost, we only performed optimal search for some of the hyperparameters. Even so, the RNN model outperformed conventional LR models. Health care systems are now investing in the infrastructure to construct some of these complex models. For example, the VHA has collaborated with Google’s DeepMind to develop an RNN model for predicting acute kidney injury using national VHA data. All deep learning neural network models including ours, have limited interpretability due to their black-box nature, which may limit acceptability by clinicians. However, recent innovations allow for interpretable deep learning models by determining the proportion of the prediction attributed to each feature.”
Assessment of a Deep Learning Model to Predict Hepatocellular Carcinoma in Patients With Hepatitis C Cirrhosis
George N. Ioannou et al.
JAMA Network Open. 2020;3(9):e2015626. Sept 2020 - "In this study, we demonstrated that RNN models that use raw longitudinal EHR data are superior to conventional LR models in estimating the risk of HCC in patients with HCV-related cirrhosis. RNN models such as ours could have multiple applications in clinical practice, provided they can be incorporated within EHR software systems.”
Assessment of a Deep Learning Model to Predict Hepatocellular Carcinoma in Patients With Hepatitis C Cirrhosis
George N. Ioannou et al.
JAMA Network Open. 2020;3(9):e2015626. Sept 2020