PSMA-PET: Deep Radiomic Biomarkers of Progression and Response Prediction in Prostate Cancer

Prostate cancer (PCa) is the most common non-skin malignancy and the third leading cause of cancer death in North American men. The accurately mapped metastatic state is a necessary prerequisite to guiding treatment in practice and in clinical trials. Imaging biomarkers (BMs) can provide information on disease volume and distribution, prognosis, changes in biologic behavior, therapy-induced changes (both responders and non-responders), durations of response, emergence of treatment resistance, and the host reaction to the therapies. Of particular relevance to metastatic prostate cancer is the emergence of a promising imaging technique involving new prostate specific membrane antigen (PSMA) positron emission tomography (PET) tracers. This approach has demonstrated higher sensitivity in detecting metastases, prior to and during therapy, than current imaging standard of care (CT and bone scan), and is not widely clinically available outside of the research realm in North America. Positron emission tomography / computer tomography (PET/CT) is a nuclear medicine diagnostic imaging procedure based on the measurement of positron emission from radiolabeled tracer molecules in vivo. PSMA is a homodimeric type II membrane metalloenzyme that functions as a glutamate carboxypeptidase/folate hydrolase and is overexpressed in PCa. PSMA is expressed in the vast majority of PCa tissue specimens and its degree of expression correlates with a number of important metrics of PCa tumor aggressiveness including Gleason score, propensity to metastasize and the development of castration resistance. \[18F\]DCFPyL is a promising high-sensitivity second generation PSMA-targeted urea-based PET probe. Studies employing second-generation PSMA PET/CT imaging in men with biochemical progression after definitive therapy suggest detection of metastases in over 60% of men imaged. Deep learning is defined as a variant of artificial neural networks, using multiple layers of 'neurons'. Deep learning