Sperm Morphology Detection & Classification

This project focuses on developing automated, deep learning-based computer vision models for sperm morphology detection and classification. The system will automatically classify sperm cells into normal and abnormal morphological categories based on WHO (World Health Organization) clinical criteria. The goal is to reduce subjectivity and inter-observer variability in manual semen analysis, which currently has up to 40% disagreement between embryologists.

• Develop deep learning models for binary classification (normal vs. abnormal sperm)
• Build multi-class classification models for morphology categorization (normal, tapered, pyriform, amorphous)
• Implement instance and semantic segmentation models to isolate sperm components (head, nucleus, acrosome, tail)
• Train and evaluate models using transfer learning approaches
• Analyze model performance using appropriate metrics (accuracy, precision, recall, IoU)
• Create interpretability visualizations (Grad-CAM, attention maps)
• Document methodology, results, and findings