STAMP FAQs

Question 1

Does STAMP provide interpretable results?

Accepted Answer

Absolutely. STAMP generates explainable heatmaps and allows for top-tile exports out-of-the-box, which are essential for transparent model auditing, scientific publications, and understanding the model's predictions.

Question 2

How does STAMP handle deep learning training with limited annotations?

Accepted Answer

STAMP utilizes end-to-end Multiple Instance Learning (MIL) with Transformer aggregation for weakly-supervised training. This means it can effectively learn from gigapixel slides using only slide-level labels, eliminating the need for time-consuming pixel-level annotations.

Question 3

What is STAMP and what is its primary purpose?

Accepted Answer

STAMP (Solid Tumor Associative Modeling in Pathology) is an end-to-end, weakly-supervised deep-learning pipeline designed to discover and evaluate candidate image-based biomarkers from gigapixel histopathology slides, without requiring pixel-level annotations.

Question 4

What kind of deep learning models does STAMP support?

Accepted Answer

STAMP offers out-of-the-box support for over 20 state-of-the-art foundation models for both tile and slide-level encoding, including popular architectures like Virchow-v2, TITAN, and COBRA.

Question 5

Is STAMP suitable for large-scale research projects?

Accepted Answer

Yes, STAMP is built for scalability. It can operate seamlessly on local machines or High-Performance Computing (HPC) environments, making it ideal for processing multi-center cohorts and extensive whole-slide image collections.

STAMP

STAMP

Key Features

Use Cases

Key Features

Use Cases