Advancing Sperm Morphology and Motility Analysis Through Artificial Intelligence

This thesis presents AI-based tools for real-time, label-free analysis of live human sperm, addressing key limitations in current clinical practices. Chapter 3 introduces the first ensemble deep learning model for classifying live sperm morphology from whole-cell images, achieving 94% accuracy benchmarked against expert annotations. Chapter 4 develops an optical flow-based tracking model, validated against both manual tracking and computer-aided sperm analysis (CASA), showing improved accuracy in key motility parameters. Chapter 5 enhances average path estimation using frequency-based smoothing method, introduces a novel metric called “path width” and applies InceptionTime to classify progressive and hyperactivated motility patterns with 89.4% accuracy.