The pathological diagnosis and treatment of azoospermia depend on precise identification of spermatogenic cells. Traditional methods are time-consuming and highly subjective due to complexity of Johnsen score, posing challenges for accurately diagnosing azoospermia. Here, we introduce a novel SC-YOLO framework for automating the classification of spermatogenic cells that integrates S3Ghost module, CoordAtt module and DCNv2 module, effectively capturing texture and shape features of spermatogenic cells while reducing model parameters. Furthermore, we propose a simplified Johnsen score criteria to expedite the diagnostic process. Our SC-YOLO framework presents the higher efficiency and accuracy of deep learning technology in spermatogenic cell recognition. Future research endeavors will focus on optimizing the model's performance and exploring its potential for clinical applications.
Keywords: SC-YOLO; YOLO framework; azoospermia; johnsen score; testicular histopathology.
YOLO framework was optimized as SC-YOLO and applied to shape detection for automatic classification of spermatogenic cells in testicular pathology images.The SC-YOLO framework has high accuracy in identifying results from binomial distributions, with or without haploid germ cells.The SC-YOLO framework also has good performance in the identification of other types of pathological sections.