PancreAI: Morphology-to-Molecular Subtype Classifier
Four morphological classes of pancreatic cancer (glandular, cribriform, solid, squamous) each have distinct transcriptomic programs and clinical implications — meaning what a pathologist sees under a microscope encodes actionable molecular information that could guide treatment decisions without…
PancreAI: Morphology-to-Molecular Subtype Classifier
Four morphological classes of pancreatic cancer (glandular, cribriform, solid, squamous) each have distinct transcriptomic programs and clinical implications — meaning what a pathologist sees under a microscope encodes actionable molecular information that could guide treatment decisions without expensive genomic testing.
Build a computational pathology tool that uses AI to classify pancreatic cancer H&E slides into the four morphological classes (glandular, cribriform, solid, squamous) and predicts the associated molecular subtype (classical vs. basal-like) and prognostic implications. The tool would be trained on publicly available PDAC tissue image datasets combined with matched transcriptomic data, allowing it to bridge the gap between histology and molecular characterization.
For clinical use, the tool would output a morphological classification with a confidence score, predicted molecular program associations, and relevant prognostic context — all derived from a standard H&E slide that every PDAC patient already has. An educational module would explain the molecular biology behind each morphological class, why solid tumors are enriched in liver metastases, and what the KRAS/EMT pathway implications are.
This tool addresses a real global health gap: molecular subtyping of PDAC requires expensive sequencing that is unavailable in most of the world's hospitals, yet H&E slides are universal. If morphological classification can serve as a reliable proxy for molecular subtype and prognosis, then a well-validated AI classifier could immediately democratize risk stratification for PDAC patients everywhere. The solid subtype's association with liver metastases alone could guide surgical decision-making.
Who Is This For?
Pathologists and oncologists in pancreatic cancer centers, particularly in resource-limited settings; computational pathology researchers developing AI diagnostic tools.
Skills & Tools Needed
- Deep learning for computational pathology (CNNs, ViTs for histology)
- Python/PyTorch or TensorFlow
- Access to PDAC histology datasets with molecular annotations (TCGA, GEO)
- Digital pathology image processing (OpenSlide, QuPath)
- Clinical knowledge of pancreatic cancer staging and treatment
Feasibility
medium — AI histopathology is an active research area with available tools and datasets, but achieving clinical-grade performance requires careful validation on large diverse cohorts — the core model is buildable but full validation is a substantial undertaking.
