Abstract

Info

Title:: An organoid segmentation method incorporating wavelet scattering and capsule network

Author(s):: YANG Hongrui1; 2; 3; LI Gang4; CHEN Zexin5; ZHAI Yujia6; 7; XU Yingying1; 2; 3; 1. School of Biomedical Engineering, Southern Medical University, Guangzhou 510515, China; 2. Guangdong ProvincialKey Laboratory of Medical Image Processing, Guangzhou 510515, China; 3. Guangdong Provincial Engineering Laboratory forMedical Imaging and Diagnostic Technology, Guangzhou 510515, China; 4. Huiqiao Medical Center, Nanfang Hospital, SouthernMedical University, Guangzhou 510515, China; 5. Guangdong Research Center of Organoid Engineering and Technology, Guangzhou510530, China; 6. Cancer Center, Affiliated Hospital of Guangdong Medical University, Zhanjiang 524001, China; 7. College ofPharmacy, Ji’nan University, Guangzhou 511436, China

Keywords:: medical image segmentation; organoid; capsule network; wavelet scattering network

Abstract:: Objective To develop and validate an automated organoid image segmentation approach based on deep learningfor addressing the issues of high misidentification rate, blurred boundary and poor generalization in current organoidsegmentation, thereby facilitating researchers to monitor and analyze organoid growth more efficiently. Methods Thewavelet scattering coefficient matrix and capsule convolution module were integrated into the U-Net architecture to constructthe organoid image segmentation model OrgCapsU-Net which was trained and evaluated on 3 organoid image datasets fromdifferent tissue sources. Results Compared with current mainstream segmentation algorithms, OrgCapsU-Net could betterdistinguish organoid and impurity, and lead to smoother segmentation boundaries, achieving superior performance across 4evaluation metrics on 3 datasets. Conclusion OrgCapsU-Net delivers excellent segmentation performance and can be appliedto organoids from various tissue sources, showing strong potential for applications in the in vitro model establishment, highthroughput drug screening, and personalized medicine.