Pathological image classification method based on diffusion attention mechanism and related device

Abstract

This invention provides a pathological image classification method and related apparatus based on a diffuse attention mechanism. The method includes: acquiring a whole-slice image of a pathological slide; preprocessing the whole-slice image to generate an original feature sequence; inputting the original feature sequence into a multi-instance learning model based on a diffuse attention mechanism for feature information interaction to obtain a target feature sequence, wherein the multi-instance learning model based on a diffuse attention mechanism employs an equidistant diffuse attention module and an attention-driven top-K-means pooling classifier; inputting the target feature sequence into an attention-driven top-K-means pooling classifier for key information extraction and aggregation to obtain a classification prediction result corresponding to the whole-slice image, wherein the attention-driven top-K-means pooling classifier employs a bypass attention mechanism. Based on this, this invention can reduce computational overhead and improve the accuracy of pathological slide image classification.

Images