Multimodal imaging and machine learning for diagnosis and patient stratification with predictive biomarkers

We propose to reduce to practice optimized multi-modal image-based (OMI) biomarkers and associated artificial intelligence/machine learning (AI/ML) software for measuring and/or predicting the responses of solid tumors to existing and investigational therapeutics. We aim to develop OMIs into  clinical tests that can deployed as Laboratory Developed Tests (LDTs) in CLIA-compliant facilities that perform diagnosis for patient care and clinical trials. Eventually we intend to test the most valuable OMI assays in clinical trials and secure FDA approval for their use as in vitro diagnostics (IVDs). Each test will be tailored to a specific type of cancer and treatment decision, providing critical information in support of precision medicine approaches. OMI tests will combine two areas of recent innovation: AI/ML modeling of conventional histological specimens, which is already being pursued commercially for diagnosis by others, and highly multiplexed tissue profiling, whose primary application is currently in research. Combining these developments into a single assay will involve a new type of microscopy we helped develop as well validated reagent panels and new AI/ML algorithms. Stage II/III colorectal cancer will serve as a test case with the goal of providing better evidence for the use of new adjuvant (post-surgery) therapies patients at highest risk of recurrence. In the course of this award we will use carefully curated retrospective patient cohorts, but future work will involve prospective populations. In proof-of-principle studies we will also test how best to use OMI spatial biomarkers in clinical trials both as measures of response and for pharmacodynamic studies. By the end of year two, the first OMI tests will be ready for deployment in HMS-affiliated institutions, and we will have a mature IP strategy to support commercialization.