Md Zobaer Islam, Ergi Spiro , Pew-Thian Yap , Michael A. Gorin and Steven P. Rowe
The diagnosis and prognosis of Prostate cancer (PCa) have undergone a significant transformation with the advent of prostate-specific membrane antigen (PSMA)-targeted positron emission tomography (PET) imaging. PSMA-PET imaging has demonstrated superior performance compared to conventional imaging methods by detecting PCa, its biochemical recurrence, and sites of metastasis with higher sensitivity and specificity. That transformation now intersects with rapid advances in artificial intelligence (AI) – including the emergence of generative AI. However, there are unique clinical challenges associated with PSMA-PET imaging that still need to be addressed to ensure its continued widespread integration into clinical care and research trials. Some of those challenges are the very wide dynamic range of lesion uptake, benign uptake in organs that may be adjacent to sites of disease, insufficient large datasets for training AI models, as well as artifacts in the images. Generative AI models, e.g., generative adversarial networks, variational autoencoders, diffusion models, and large language models have played crucial roles in overcoming many such challenges across various imaging modalities, including PET, computed tomography, magnetic resonance imaging, ultrasound, etc. In this review article, we delve into the potential role of generative AI in enhancing the robustness and widespread utilization of PSMA-PET imaging and image analysis, drawing insights from existing literature while also exploring current limitations and future directions in this domain.