การพัฒนาขั้นตอนวิธีแยกส่วนไวต์พัลป์ สําหรับข้อมูลไครโอ อิมเมจจิงของม้ามหนู

Abstract

Spleen is one of the organs of interest regarding the study of the prognosis of graft-versushost disease (GVHD) via cryo-imaging. Until now, there has yet to be a segmentation algorithm developed for cryo-images of spleen tissue. In this research, we developed a new algorithm that was able to segment cryo-images of spleen tissue automatically. The algorithm consisted of (1) preprocessing, (2) predicting, and (3) post-processing. Two U-Nets were used, one for separating spleen from background, and another for separating white pulp from red pulp. The results created by algorithm closely resembled the ground truth created by cryo-imaging experts. The Dice similarity coefficient was approximately 82%. We also measured the white pulp volume to spleen volume ratio. The results showed that the ratio of the group with GVHD was significantly higher than the group without GVHD, which was consistent with the original study. In conclusion, the algorithm was able to segment spleen tissue with accuracy and quality near that of the experts. Once the algorithm is incorporated into the T-cell proliferation assay workflow, we expect that it will greatly increase throughput of the process as compared to manual segmentation by humans, and in turn help accelerate the research for further understanding of graft-versus-host disease and development of the cure.