STAMP

Acerca de

STAMP (Solid Tumor Associative Modeling in Pathology) is an efficient, ready-to-use workflow designed for computational pathology. It provides an end-to-end, weakly-supervised deep-learning pipeline that enables clinical researchers and machine-learning engineers to discover and evaluate candidate image-based biomarkers directly from gigapixel whole-slide histopathology images, without requiring pixel-level annotations. The tool supports scalable execution on local machines or HPC environments, integrates over 20 foundation models, and generates explainable heatmaps and metrics for reproducible research and biomarker prediction.

Características Principales

• Generates explainable heatmaps and top-tile exports for model auditing and publication
• Scalable operation on local machines or High-Performance Computing (HPC) with a unified CLI
• Compatible with Model Context Protocol (MCP) for integration into agentic AI workflows
• Supports over 20 foundation models for tile and slide level encoding (e.g., Virchow-v2, TITAN, COBRA)
• 84 GitHub stars
• Utilizes end-to-end Multiple Instance Learning (MIL) with Transformer aggregation for weakly-supervised training

Casos de Uso

• Predicting patient survival directly from H&E whole-slide images in squamous carcinoma studies.
• Serving as an open-source framework for standardized Whole-Slide Imaging (WSI) tiling and feature extraction in foundation model benchmarking and evaluation.
• Accurately predicting histologic disease activity scores from H&E tissue sections in inflammatory bowel disease studies.