Deep Learning-Based Image Analysis Model for Classification and Quantification of Multiple Histopathological Findings...

Area of Science:

• Veterinary Pathology
• Computational Pathology
• Toxicology

Background:

• Supervised deep learning models show promise for analyzing histopathological findings in laboratory animals.
• Existing models struggle to simultaneously detect multiple abnormal testicular findings in rat H&E-stained specimens.

Purpose of the Study:

• To develop a deep learning model for simultaneous detection, classification, and quantification of major testicular toxicity findings in rats.
• To classify spermatogenic stages on H&E-stained whole slide images (WSIs).

Main Methods:

• Utilized supervised deep learning algorithms and WSI datasets for model training.
• Trained the model on WSIs of rat testes and epididymis from toxicity studies.
• Included 7 key testicular toxicity findings: germ cell degeneration, tubular atrophy, tubular dilatation, Sertoli cell vacuolation, multinucleated giant cells, and epididymal sperm/debris reduction.

Main Results:

• Achieved high detection accuracy for classifying spermatogenic stages and the 7 identified testicular findings.
• Model performance was validated against diagnoses from board-certified pathologists.
• Demonstrated high performance in both classification and quantification tasks.

Conclusions:

• The developed model serves as a valuable tool to support histopathological evaluation in rat toxicity studies.
• It is particularly useful for initial screening, enhancing work efficiency.
• The model is expected to reduce errors by preventing oversights in histopathological assessments.