Katherine R Seymour, Jessica P Rickard, Simon P de Graaf
Sperm morphology is a critical predictor of fertility across a range of species; however, its assessment remains inherently subjective and thus susceptible to variability and inaccuracy. Morphological abnormalities can present in diverse forms, with over 25 distinct categories recognised in the current literature. These abnormalities may arise during spermatogenesis due to stressors, as well as post-maturation during epididymal storage, ejaculation, or as a consequence of suboptimal handling during processing. Researchers have long sought to characterise these abnormalities through classification systems. Numerous classification frameworks have been developed to categorise observed defects. However, inconsistencies in the classification of specific abnormalities have led to disputes among experts. Preferences for particular classification schemes vary among laboratories, species, industries, and geographic regions, resulting in a lack of standardisation that hampers comparability and contributes to considerable inter- and intra-observer variability. While advancements in automation have begun to emerge, subjective visual assessment remains the predominant method of evaluation. This reliance, combined with inconsistent classification standards, absence of formal training, and inherent human bias, continues to drive substantial variability. In response, recent research has turned to novel technologies such as machine learning to achieve objective, scalable assessment. However, training these models necessitates high-quality training data -something difficult to obtain given the challenges of standardising subjective assessments. To preserve the utility of sperm morphology as a useful indicator of fertility, it is essential that future efforts focus on improving assessor training and standardisation protocols, thereby reducing variability while the development of automated, objective technologies continues to progress.