Multi-Axis Attention with Convolution Parallel Block for Organoid Segmentation

Abstract

Dear Editor, This letter presents an organoid segmentation model based on multi-axis attention with convolution parallel block. MACPNet adeptly captures dynamic dependencies within bright-field microscopy images, improving global modeling beyond conventional UNet. It excels in sparse global interactions and concurrent computation, yielding enhanced segmentation. MACPNet stands out for its prowess in multi-scale data capture, aligned with diverse distance dependencies inherent in organoid images. Experimental results show that the proposed model outperforms several state-of-the-art methods as well as multiple baseline models in accurate organoid segmentation.