CaCL: Class-Aware Codebook Learning for Weakly Supervised Segmentation on Diffuse Image Patterns

Weakly supervised learning has been rapidly advanced in biomedical image analysis to achieve pixel-wise labels (segmentation) from image-wise annotations (classification), as images naturally contain many scenarios. The current weakly algorithms the computer vision community are largely designed for focal objects (e.g., dogs and cats). However, such not optimized diffuse patterns imaging stains fluorescence microscopy imaging). In this paper, we propose a novel class-aware codebook (CaCL) algorithm perform patterns. Specifically, CaCL is deployed segment protein expressed brush border regions histological of human duodenum. Our contribution three-fold: (1) approach segmentation perspective; (2) segments rather than objects; (3) proposed implemented multi-task framework based on Vector Quantised-Variational AutoEncoder (VQ-VAE) via joint reconstruction, classification, feature embedding, segmentation. experimental results show that our method achieved superior performance compared with baseline algorithms. code available at https://github.com/ddrrnn123/CaCL.