AI System for Analyzing Medical Images Using Magnification-Aligned Transformers

This 2025 patent describes an AI system that analyzes medical images, like those from pathology slides, by using a special 'Transformer' model that understands how image details change with magnification.

What it covers

This patent details a method for analyzing pathological images, such as those from whole-slide images (WSIs) used in medical diagnosis. First, it identifies and isolates the actual tissue regions within these large images, discarding blank or unusable areas. Then, it breaks down these tissue regions into smaller 'patches.' The core of the invention is a 'Magnification-Aligned Transformer' (MAT) classification network model. This model has two main parts: one that aligns features from images taken at different magnifications (like zooming in or out) and another that uses both global (big picture) and local (detailed) attention mechanisms, combining the strengths of Transformers for overall understanding and Convolutional Neural Networks (CNNs) for fine details. This trained model then predicts a classification for the pathological image, helping doctors diagnose conditions. For example, it could help identify cancerous cells in a tissue sample.

What it doesn't cover

• —Analyzing images that are not pathological or medical in nature.
• —Methods that do not involve breaking down whole-slide images into patches.
• —AI models that do not use a Transformer architecture.
• —Systems that do not specifically align features across different image magnifications.
• —Analysis that relies solely on global features without considering local details.

The clever bit

The clever part is how the system aligns features extracted from images at different magnifications. It uses a self-supervised approach to ensure that the AI understands the same underlying tissue structures whether it's looking at a zoomed-out overview or a highly magnified view, preventing loss of critical detail during analysis.

Why it matters

This patent is significant because it addresses the challenge of analyzing complex medical images, particularly whole-slide pathology images, which are massive and contain crucial diagnostic information at various levels of detail. By developing an AI that can intelligently process these images across different magnifications, it aims to improve the accuracy and efficiency of disease diagnosis, potentially aiding pathologists in identifying subtle abnormalities.

Real-world examples

• 1.AI-assisted pathology slide analysis software
• 2.Digital pathology platforms
• 3.Computer-aided diagnosis systems for cancer detection