Multimodal AI · Prostate Cancer

Smarter Risk Stratification for Active Surveillance

A lightweight, interpretable AI engine that combines pathology, MRI, PSA, and clinical data to identify which men on active surveillance are at highest risk of adverse reclassification.

The Clinical Problem

Active Surveillance Needs Better Decision Support

Roughly half of men diagnosed with low-risk prostate cancer are managed with active surveillance — watchful waiting rather than immediate treatment. But current tools cannot reliably identify which patients will progress to clinically significant disease, leading to either over-treatment or missed windows for timely intervention.

~50%

of newly diagnosed prostate cancers managed with active surveillance

30–40%

of active surveillance patients reclassify to higher-risk disease within 5 years

Single-modal

current tools rely on pathology or PSA alone — missing the full clinical picture

Our Approach

A Multimodal Risk Engine, Not Just an Image Classifier

We combine expert-defined quantitative pathology features from whole-slide imaging with MRI findings, PSA density, clinical variables, and longitudinal change signals — producing a calibrated risk score that is interpretable by clinicians and actionable at the point of care.

01

Pathology Intelligence Layer

Expert GU pathologists prospectively define clinically meaningful morphologic variables from whole-slide images — anchoring the model in interpretable biology rather than opaque embeddings.

02

Multimodal Fusion

Pathology features are combined with MRI findings, PSA/PSA density, clinical factors, and longitudinal change signals to capture the full risk picture.

03

Lightweight & Interpretable

The model uses structured, clinically meaningful features — not a massive foundation model — making it fast, auditable, and deployable in real clinical workflows.

04

Expandable Platform

The same infrastructure supports future outputs: adverse pathology at prostatectomy, biochemical recurrence prediction, and treatment-intensification risk scoring.

Why This Is Different

Expert-Guided Pathology as a Competitive Moat

The defensible differentiation is not institution versus institution. It is the specific asset stack: expert-defined quantitative pathology combined with a multimodal clinical endpoint, developed by a multidisciplinary NYU team with an existing TOV technology disclosure.

NYU-Disclosed Technology

An existing NYU Technology Opportunities & Ventures disclosure provides a meaningful commercialization foundation and IP posture.

Multidisciplinary Development

GU pathology, urology, radiation oncology, medical oncology, AI engineering, and biomedical informatics — all contributing to a single clinical decision.

Versioned, Reproducible Pipeline

Pathology feature extractors are versioned and frozen at stability milestones, enabling rigorous validation and transparent model updates.

Near-Term Milestones

Building Toward Clinical Validation

01

Confirm Clinical Co-Investigator

Expert NYU urologist confirmed as clinical co-investigator; initial population and primary outcome defined.

02

Pathology Feature Dictionary

First expert-defined quantitative morphology dictionary specified, including conventional pathology variables and optional learned representations.

03

WSI Feature Stability

Objective technical criterion for whole-slide image feature stability defined at the ~800 WSI stage; v1 feature extractor frozen.

04

Multimodal Prototype

Baseline clinical-only model built, then pathology + clinical, then MRI + clinical, then full multimodal — quantifying incremental value at each step.

05

Validation Framework

Patient-level temporal and external validation planned from the start; calibration and provenance tracking built into the pipeline.

Interested in Collaborating or Investing?

We are actively seeking clinical co-investigators, AI engineering collaborators, and early-stage investors aligned with our commercialization pathway.

ProstatAI

A multimodal AI risk engine for prostate cancer active surveillance — combining pathology, MRI, PSA, and clinical data to support smarter clinical decisions.

NYU Research

Connect

This platform is an investigational research tool and has not been approved by the FDA or any regulatory authority. Information on this site is intended for research and educational purposes only and does not constitute medical advice or clinical guidance.

© 2026 NYU Multimodal Prostate AI Project. All rights reserved.