Research News
AI-Based Method Accurately Segments and Quantifies Overlapping Cell Membranes

Researchers at University of Tsukuba have developed DeMemSeg, an AI-powered analysis pipeline that addresses a long-standing challenge in microscopy: precisely segmenting and measuring individual cell membranes that overlap in two-dimensional (2D) projection images. This innovation is expected to accelerate research on cellular mechanisms and related diseases.
Tsukuba, Japan—The shape of cells and their organelles is closely tied to their function, making accurate measurement essential for understanding fundamental biological processes. Biological analysis often requires projecting three-dimensional (3D) fluorescence images into 2D for visualization. However, during this process, structures that are distinct in 3D frequently appear to overlap in 2D, complicating the accurate segmentation of individual contours.
To overcome this issue, researchers developed DeMemSeg, a novel AI-driven pipeline based on deep learning. Using the prospore membrane formed during sporulation in budding yeast as a model system, the team showed that DeMemSeg can automatically segment overlapping membrane structures with accuracy statistically comparable to that of expert manual analysis. Moreover, when applied to mutant cells with abnormal membrane morphologies not represented in its training set, DeMemSeg successfully detected and quantified these irregularities, demonstrating a strong generalization ability.
This approach enables large-scale, objective, and quantitative analysis of morphological data that was previously difficult to obtain, especially in budding yeast research. In addition, the workflow underlying DeMemSeg can be adapted to other areas of life sciences. Because certain human disorders of gamete formation and fertilization involve morphological abnormalities in cells or organelles, these findings provide a foundational technology for advancing the understanding of such disease mechanisms.
###
This research was supported by JSPS KAKENHI Grant Number 22K06074 (to KI), and Ohsumi Frontier Science Foundation (to YS).
Original Paper
- Title of original paper:
- Deep Learning-Based Segmentation of 2D Projection-Derived Overlapping Prospore Membrane in Yeast
- Journal:
- Cell Structure and Function
- DOI:
- 10.1247/csf.25032
Correspondence
Associate Professor SUDA Yasuyuki
Institute of Medicine, University of Tsukuba
TAGUCHI Shodai
Ph.D. Program in Humanics, School of Integrative and Global Majors, University of Tsukuba