Prostate Cancer Risk Stratification Using Needle Core Biopsy or Radical Prostatectomy Specimens (Continuation of PLCOI-2074)
Principal Investigator
Name
Uttara Joshi
Degrees
MBBS, DNB (Pathology)
Institution
AIRA Matrix Private Limited
Position Title
Pathologist, Head -Strategic Initiatives
About this CDAS Project
Study
PLCO
(Learn more about this study)
Project ID
PLCOI-2080
Initial CDAS Request Approval
Sep 8, 2026
Title
Prostate Cancer Risk Stratification Using Needle Core Biopsy or Radical Prostatectomy Specimens (Continuation of PLCOI-2074)
Summary
AI-based digital pathology models have shown increasing promise for extracting prognostically relevant morphological features from H&E-stained whole-slide images (WSIs) that are not readily captured by conventional grading alone. AIRA Matrix has developed such a model, AIRAStrat-Prostate, which analyzes H&E whole-slide images from prostate needle core biopsy and radical prostatectomy specimens using a locked, validated deep learning pipeline to generate an individualized prognostic risk score and corresponding risk category (low, intermediate, high) for distant metastasis, alongside an explainability heatmap highlighting the image regions contributing most to the model's output.
Before such a model can be responsibly considered for broader use, it must be validated in patient cohorts independent of its original development set, ideally spanning diverse practice settings, demographics, and follow-up conditions, and across both specimen types on which it may be applied. The PLCO Cancer Screening Trial offers a uniquely valuable validation resource: a large, prospectively followed cohort of men with prostate cancer, linked clinical, diagnostic, and treatment data (including specific identification of needle core biopsy and radical prostatectomy), long-term mortality follow-up with adjudicated cause of death, and a substantial bank of digitized H&E whole-slide prostate pathology images already scanned and available through the Cancer Genomics Research Laboratory (CGR). This project proposes to use these resources to independently validate AIRAStrat-Prostate's prognostic and predictive performance in both the PLCO biopsy and radical prostatectomy cohorts.
Aims
Aim 1: Investigate the performance of validated prognostic and predictive AI algorithms for men undergoing prostate needle core biopsy and/or radical prostatectomy
We will apply the locked AIRAStrat-Prostate model to digitized H&E whole-slide images from the PLCO cohort, separately for men with a diagnostic needle core biopsy and men who underwent radical prostatectomy and assess the model's discrimination and calibration for predicting disease recurrence and distant metastasis, using PLCO's linked diagnostic, treatment, and mortality data to define outcomes. This constitutes an independent, external validation of the model, across both specimen types, in a cohort distinct from its original development and internal validation sets.
Aim 2: Explore the prognostic performance of AI algorithms in clinically-relevant patient subpopulations. We will evaluate whether the model's prognostic performance is consistent across clinically meaningful subgroups, including specimen type (biopsy versus RP), pathologic/biopsy stage, Gleason grade group, age at diagnosis, race/ethnicity, and screening arm assignment, in order to identify any subpopulations where performance may be attenuated or enhanced, and to characterize the model's generalizability across the diversity represented in the PLCO cohort.
Aim 3: Compare performance of AI algorithms to established clinical risk stratification tools. We will benchmark the AI model's discrimination and net clinical benefit against established risk stratification tools that can be constructed from PLCO's available clinical and pathologic variables, CAPRA-like composite scores at the biopsy stage, and Gleason grade group, pathologic stage-based grouping, and CAPRA-S-like composite scores at the post-surgical stage to characterize whether the AI model provides incremental prognostic value beyond current standard-of-care tools at either decision point.
Collaborators
Uttara Joshi AIRA Matrix Private Limited
Nitin Singhal AIRA Matrix Private Limited
Tamara Lotan Johns Hopkins School of Medicine