Deep learning-based segmentation of 2D projection-derived overlapping prospore membrane in yeast

Area of Science:

• Cell Biology
• Biophysics
• Computational Biology

Background:

• Quantitative morphological analysis is vital for understanding cellular processes.
• 3D Z-stack imaging provides high-resolution data but poses challenges for manual annotation and interpretation due to complex 3D structures.
• Maximum Intensity Projection (MIP) simplifies 2D visualization but causes artificial overlaps, hindering automated segmentation of individual structures.

Purpose of the Study:

• To develop an automated deep learning pipeline for segmenting overlapping membrane structures in 2D MIP images.
• To address the limitations of conventional methods in analyzing complex, overlapping cellular morphologies.
• To enable accurate quantitative morphological analysis from widely used 2D projection images.

Main Methods:

• Developed DeMemSeg, a deep learning pipeline based on Mask R-CNN.
• Trained DeMemSeg on a custom-annotated dataset using a systematic image processing workflow.
• Validated the model's performance against expert manual annotation and on unseen data from gip1Δ mutant cells.

Main Results:

• DeMemSeg accurately identifies and delineates individual, overlapping prospore membranes (PSMs) in yeast sporulation.
• Segmentation performance and derived morphological measurements were statistically indistinguishable from expert manual annotation.
• The model successfully generalized to segment PSMs in gip1Δ mutant cells, capturing morphological defects.

Conclusions:

• DeMemSeg provides a robust, automated solution for objective quantitative analysis of complex, overlapping membrane morphologies.
• The pipeline enables accurate analysis directly from 2D MIP images, overcoming limitations of conventional segmentation methods.
• DeMemSeg offers an adaptable workflow to advance cell biology research by facilitating detailed morphological studies.