Deep learning enabled multi-organ segmentation of mouse embryos

Area of Science:

• Developmental biology
• Bioinformatics
• Medical imaging

Background:

• The International Mouse Phenotyping Consortium (IMPC) provides extensive 3D mouse embryo imaging data.
• Manual segmentation of these 3D images is computationally intensive and time-consuming, hindering research.
• Automated tools are needed to facilitate the analysis of phenotype/genotype interactions from this valuable data.

Purpose of the Study:

• To develop and validate an open-source, deep learning-based tool for segmenting multiple anatomical structures in 3D mouse embryo images.
• To provide researchers, including those without coding experience, with an accessible platform for analyzing complex imaging data.
• To streamline the process of investigating genotype-phenotype relationships using high-resolution embryonic imaging.

Main Methods:

• Development of Mouse Embryo Multi-Organ Segmentation (MEMOS), a deep learning tool integrated with the 3D Slicer platform.
• MEMOS automatically segments approximately 50 anatomical structures within 3D mouse embryo images.
• The tool includes functionalities for manual review, editing, and analysis of segmentations within a single application.

Main Results:

• MEMOS successfully estimates segmentations for 50 anatomical structures in 3D mouse embryo data.
• Performance validation demonstrates comparable results to state-of-the-art atlas-based segmentation methods.
• The tool enabled the quantification of anatomical abnormalities in a Cbx4 knockout mouse strain, confirming its utility.

Conclusions:

• MEMOS offers an accessible and efficient solution for the segmentation of 3D mouse embryo images.
• The tool significantly reduces the barrier to entry for analyzing IMPC data, facilitating genotype-phenotype research.
• MEMOS empowers researchers to explore complex developmental phenotypes through automated and interactive image analysis.