Morpho-functional analysis of bovine spermatozoa using cytometry with deep learning algorithms

Abstract

This study presents a high-throughput approach for single-cell morpho-functional analysis of bovine spermatozoa by integrating imaging flow cytometry with deep learning. A dataset of over 400,000 sperm images was used to train a ConvNeXt-Tiny convolutional neural network for automated classification into ten morphological categories. Functional analysis was performed on approximately 60,000 images obtained from cryopreserved semen, resulting in a final dataset of 40,613 high-quality single-cell events annotated with mitochondrial membrane potential (MMP) and intracellular calcium levels, stained with tetramethylrhodamine ethyl ester perchlorate (TMRE) and Fluo-4, respectively. Statistical analysis using the Kruskal–Wallis test revealed significant differences in calcium levels across morphological groups (p < 0.001), with post-hoc Dunn’s test identifying multiple group-level differences. Results demonstrated that abnormal midpiece, coiled tail, and distal cytoplasmic droplet groups exhibited distinct functional profiles, including altered mitochondrial activity and calcium signaling. This study represents the first large-scale morpho-functional analysis of spermatozoa from Kazakhstani cattle breeds at single-cell resolution. The developed pipeline provides a foundation for improved fertility assessment and highlights the potential of AI-driven multi-parametric analysis in reproductive biology.