Single-cell three-dimensional tracking by means of neural networks for sperm rolling classification.

Abstract

Single-cell analysis enables the extraction of detailed information from individual cells that bulk analysis cannot provide. Convolutional neural networks (CNNs) hold great potential in overcoming the challenges deriving from single-cell tracking, providing a powerful framework for automated and high-throughput analysis. In this field, sperm analysis for male infertility assessment finds its application. Indeed, evaluating sperm quality indicators of motility and morphology is essential for this purpose, although the gold standard analysis still relies on manual assessment. Here, we propose an automated, label-free method for sperm rolling detection and analysis based on CNNs. Brightfield image sequences of swimming sperm are captured with the same magnification for both motility and morphology analysis. This workflow is based on sperm head detection, identifying-for the first time-the three-dimensional configuration assumed during the motion. Following steps of tracking and segmentation enable the simultaneous extraction of kinematic and morphometric parameters from the head contour across frame sequences, providing additional information related to sperm rolling. The approach successfully captures motion changes, demonstrating its ability to perform advanced sperm characterization. Correlating kinematics and morphology at the single-cell level, the proposed method enhances insights into motility and provides more accurate sperm characterization.