Research Projects

Prostate cancer (or PCa) is the third most common cancer among males in India and is estimated at 7% of all cancer incidences reported nationally in 2022. It is majorly diagnosed in aging men, with the average age of diagnosis being around 61, and as per the NCRP reports, the PCa burden will be substantially increased in India by 2025. While, prostate-specific antigen (PSA) serum, physical examination (DRE), and multi-parametric magnetic resonance imaging (mpMRI) have been the primary standard-of-care detection and quantifying responses to onset treatments, histology is currently accepted as a gold standard for pathologic analysis of PCa evaluated in surgical specimen. However, it is still challenging to predict treatment response by stratifying underlying risk and associating the pathological Gleason grades with clinical TNM staging using baseline and/or follow-up records. Despite the availability of standard diagnostic protocols and screening modalities, there is an absence of definitive markers for detecting early-grade tumors, and inter-modality disagreements in stage evaluation, recurrence and survival prediction to pose bottlenecks for clinicians. Thus, artificial intelligence (AI)-driven companion diagnostics and decision support systems are crucial in improving clinical procedures, risk predictions, and overall outcomes. The key challenge is poor specificity and sensitivity in identifying the lower-risk and higher-risk patients based on clinical/pathological grades of tumor using baseline, pre-and post-operative data, resulting in misdiagnosis or false/over-treating patients. Therefore, there is a critical need for identifying novel computational markers on structural images and their concordant association to pathological findings in prostate cancers, which could have a significant clinical impact in identifying patients' eligibility for personalized care thus providing improved mortality rate and quality of life. The utilization of AI notably improves the accuracy and efficiency of oncologic research, opening doors to personalized cancer treatments by integrating multi-domain data into the framework. In this study, the main objective is to discover novel markers using clinical-radiological-pathological domain fusion (cRadPath-Fuse) to accurately identify the patients in different response groups using the ML/DL framework. The cRadPath-Fuse model will leverage new structural diversity descriptors for spatial heterogeneity and morphometric features on MRI and WSI, to provide predictive analysis. The dataset will be retrospectively collected from different sites and images will be annotated by the expert radiologists, and pathologists before the next pre-processing steps. Further, the training and testing will be done using ML/DL algorithms. The validation data collected from the different sites will be used to justify the variability and reproducibility of the model.