Overview
We use AI-enabled pathology and multimodal data integration to connect tissue morphology, molecular features, and clinical outcomes in pancreatic cancer.
Approach
Computational pathology workflows can help quantify tumor architecture, stromal context, immune infiltration, and treatment-associated tissue states.
These image-derived features can be integrated with genomics, therapeutic response, and clinical annotations to identify translational hypotheses for prospective study.
Multimodal integration
Whole-slide images, multiplexed tissue imaging, spatial transcriptomics, and bulk genomics are aligned on the same specimens so that morphological patterns can be traced back to molecular drivers.
Deep-learning models are trained with careful attention to batch effects, site heterogeneity, and validation on held-out cohorts to keep image-derived biomarkers reproducible and clinically credible.
Why it matters
Tissue is the richest, most routinely available record of a pancreatic tumor. Reading it computationally lets us surface prognostic and predictive signals that complement sequencing — and do so from the biopsy a patient already has.