Deep Learning-Based Spermatogenic Staging in Tissue Sections of Cynomolgus Macaque Testes

Area of Science:

• Reproductive toxicology
• Computational pathology
• Primate reproductive biology

Background:

• Assessing male reproductive toxicity requires evaluating testicular histology.
• Microscopic staging of seminiferous tubules is crucial but subjective and labor-intensive.
• Digital pathology and AI offer potential for automating complex histological assessments.

Purpose of the Study:

• To develop and validate a deep learning model for automated staging of seminiferous tubules in cynomolgus monkey testes.
• To improve the efficiency and objectivity of fertility adverse effect evaluations.
• To explore the utility of stage-frequency maps derived from automated analysis.

Main Methods:

• Development of a deep learning model utilizing digital whole slide images (WSIs) of testis tissue sections.
• Validation of the model on six WSIs using a six-stage spermatogenic classification system.
• Quantitative analysis of the model's performance in staging an average of 4938 seminiferous tubule cross-sections per WSI.

Main Results:

• The deep learning model achieved high sensitivity, precision, and accuracy in annotating tubule stages.
• On average, 78% of seminiferous tubules were accurately staged across the WSIs.
• The model facilitated the creation of stage-frequency maps, detailing the distribution of spermatogenic stages.

Conclusions:

• Automated deep learning analysis of testicular WSIs can accurately stage seminiferous tubules.
• This AI-driven approach supports pathologists in stage-aware evaluations and may aid in reproductive toxicology studies.
• Further research is needed to establish the diagnostic value of stage-frequency maps, particularly in cases of spermatogenic disturbances.