Sperm morphology is a critical predictor of fertility in both livestock and human assisted reproduction (1). However, it remains a subjective assessment, vulnerable to human bias and variability unless morphologists adhere to rigorous standardisation protocols. Despite this, there is currently no standardised training tool for sperm morphology assessment, contributing to significant inter- and intra-variation amongst morphologists.
This study developed (2) and evaluated (3) the effectiveness of a novel sperm morphology assessment training tool underpinned by machine learning principles. Specifically, the concept of supervised learning from expert-labelled “ground truth” datasets. The tool presents users with sperm images individually classified by multiple experts and provides immediate feedback on classification accuracy.
Using the tool, novice users were trained to recognise 25 different morphological sperm types and then assessed on their ability to categorise these into four classification systems of increasing complexity (2-, 5-, 8-, and 25-category). Untrained users (n=22) achieved modest accuracy (81.0 ± 2.5%, 68 ± 3.59%, 64 ± 3.5%, and 53 ± 3.69% across the four systems, respectively), while users exposed to a training video and visual aid (n=16) displayed significantly higher accuracy in their initial tests (94.9 ± 0.66%, 92.9 ± 0.81%, 90 ± 0.91% and 82.7 ± 1.05, p < 0.001). Continued weekly training further increased accuracies for this latter cohort to 98 ± 0.43%, 97 ± 0.58%, 96 ± 0.81%, and 90 ± 1.38% (p<0.001).
Although machine learning may ultimately automate sperm morphology assessment, this study demonstrates that applying its training principles to human users via high-quality, expert-validated image sets and structured, feedback-driven repetition can greatly enhance human diagnostic performance in the interim. This tool represents a scalable, standardised method to improve reproducibility and accuracy in sperm morphology assessment and may serve as a foundation for broader implementation across species and classification systems.