SATINN v2: automated image analysis for mouse testis histology with multi-laboratory data integration†

Area of Science:

• Reproductive Biology
• Computational Pathology
• Bioinformatics

Background:

• Testis histology analysis is crucial for male fertility studies but is time-consuming and requires specialized expertise.
• Automating this process can significantly improve efficiency and accessibility.

Purpose of the Study:

• To develop and validate novel neural network models for automated classification of cell types and tubule stages in mouse testis histology images.
• To assess the generalizability and robustness of the developed tool across different datasets and laboratories.

Main Methods:

• Development of neural network models for automated classification of 14 cell types and 12 tubule stages from whole-slide brightfield images.
• Extensive training diversification and testing on external wildtype and mutant datasets from multiple laboratories.
• Application of the tool (SATINN) to analyze testis images from eight mutant lines and cluster histology images in latent space.

Main Results:

• The cell type classifier achieved 96% external validation accuracy.
• The tubule stage classifier achieved 63% accuracy (96% with ±1 stage tolerance).
• SATINN successfully processed diverse datasets and revealed known pathological relationships among mutant lines when applied to latent space clustering.

Conclusions:

• SATINN represents a significant advancement in automated testis histopathology, offering a robust tool for multi-laboratory use.
• The tool enhances the analysis of male fertility and associated pathologies, facilitating broader research applications.