An automatic classification method of testicular histopathology based on SC-YOLO framework

Area of Science:

• Reproductive Biology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate identification of spermatogenic cells is crucial for diagnosing azoospermia.
• Traditional methods, like the Johnsen score, are subjective and time-consuming.
• Existing diagnostic challenges necessitate advanced automated solutions.

Purpose of the Study:

• To develop and validate a novel deep learning framework (SC-YOLO) for automated spermatogenic cell classification.
• To improve the accuracy and efficiency of azoospermia diagnosis.
• To simplify diagnostic criteria for faster clinical application.

Main Methods:

• Developed the SC-YOLO framework integrating S3Ghost, CoordAtt, and DCNv2 modules.
• Employed deep learning for automated classification of spermatogenic cells.
• Proposed simplified Johnsen score criteria for expedited diagnosis.

Main Results:

• The SC-YOLO framework demonstrated high efficiency and accuracy in spermatogenic cell recognition.
• The integrated modules effectively captured essential texture and shape features.
• Reduced model parameters while maintaining performance.

Conclusions:

• The SC-YOLO framework offers a significant advancement in automated spermatogenic cell recognition.
• Deep learning provides a more efficient and accurate alternative to traditional diagnostic methods.
• Future work will focus on clinical validation and model optimization.