Artificial intelligence in male infertility from diagnosis to treatment

Abstract

Male factors account for almost half of all infertility cases and afflict approximately 186 million men globally, presenting a substantial and increasing public health burden. Semen analysis is the standard of care for assessing male fertility, but subjective visual interpretation results in > 25–30% inter-observer variability and precludes non-destructive assessment of sperm DNA fragmentation (SDF). Machine learning, a branch of artificial intelligence, has the potential to streamline semen assessment and clinical decision-making throughout male infertility care by providing automated, data-driven approaches that are objective, efficient, and precise. Guided by the PRISMA 2020 statement, the manuscript included research from PubMed, Scopus, and Google Scholar. Two independent reviewers screened all retrieved records according to pre-specified inclusion criteria to identify peer-reviewed publications describing applications of machine learning, deep learning, artificial neural networks, natural language processing, computer vision, and explainable artificial intelligence to semen analysis, fertility status prediction, or assisted reproductive technologies. Artificial Intelligence techniques consistently outperformed conventional semen assessment methods for sperm count, motility, morphology, and predictive analyses. Feedforward Spectral Neural Networks classified sperm concentration with 93% accuracy, while a Fast Region-based Convolutional Neural Network classified sperm motility with 97.37% accuracy; both surpassed inter-observer agreement achievable through WHO-based manual visual analysis. VGG16-pretrained CNNs identified sperm morphology with a true positive rate of 94.1%, however, performance declined substantially to 62% when validated on an independent external dataset. Deep learning models demonstrated the capacity to predict SDF non-destructively, high concordance to TUNEL assay and sperm chromatin structure assay reference standards (r = 0.97, p < 0.001). Machine Learning classifiers specifically, Extreme Gradient Boosting (XGBoost), Random Forest, and CatBoost, showcased AUCs of 0.998 for the prediction of male fertility status. Furthermore, explainable artificial intelligence techniques such as SHAP and LIME improve the clinical interpretability of machine learning model outputs, an important step toward regulatory acceptance. These techniques can also be integration with computer-assisted sperm analysis systems, smartphone-based semen analysis, sperm sorting for intrauterine insemination, microTESE, ultrasonography, and multi-omics data platforms. Limitations of the current literature include retrospective study designs, single-centre data sources, small sample sizes, lack of external validation, and heterogeneity in image acquisition, outcome measures, and performance metrics. Included studies face challenges relating to algorithmic bias, insufficient model explainability, data security and governance, and the absence of standardized regulation or AI-specific reporting frameworks. Future work should prioritise prospective multicentre studies employing multimodal artificial intelligence that integrates imaging data with genomic, transcriptomic, hormonal, and electronic health record inputs, as well as federated learning and international collaboration to establish standardised and internationally recognised regulatory guidelines. In conclusion, artificial intelligence has the potential to transform clinical decision-making for patients presenting with infertility by enabling objective, high-throughput, and personalised assessment of semen samples. However, these tools should be used to augment rather than replace clinical expertise until models have been prospectively validated across diverse populations and have received appropriate regulatory clearance. Not applicable.