High-throughput label-free assessment of sperm DNA fragmentation index via intelligent morphological imaging

correlation analysis. Convolutional neural networks are then employed to extract deep learning features for enhanced classification. Following measurements of 31 clinical semen samples and analysis of the resulting 136 070 images, the classification accuracy for individual sperm with low, medium, and high DFI is 82.61%, 80.39%, and 82.06% respectively, while sample-level classification achieves complete agreement with clinical tests through group-based majority voting mechanisms. Furthermore, we establish a quantitative comprehensive score metric integrating classification proportions across DFI groups, enabling continuous numerical assessment. This score shows strong concordance with clinical DFI values and closer consistency with conventional semen parameters. We believe that this work provides an intelligent, high-throughput, label-free sperm DFI assessment method, demonstrating potential as a solution for clinical diagnosis of male infertility.