A method and device for colon polyp image segmentation based on prompt guidance

Abstract

The application provides a colon polyp image segmentation method and device based on prompt guidance, wherein the method comprises the following steps: a plurality of multi-scale image blocks are generated from an original image according to prompt information through a sampling operator; then, a local semantic feature of the image block is extracted by an encoder, and an inductive bias is introduced; a long-range dependence between the image blocks is modeled and global information is aggregated by a Transform module; the feature is reconstructed step by step by a decoder, and the spatial resolution is recovered; finally, the predicted image block is recombined to the original spatial position through an inverse operator, and a full-resolution segmentation result is obtained. The application can effectively focus on small-size, irregular and edge-blurred difficult segmentation areas, significantly improve the segmentation accuracy and robustness of the model, and has good clinical application potential and popularization value.

Images