Artificial Intelligence for Clinical Management of Male Infertility, a Scoping Review

Abstract

Purpose of review: Infertility impacts one in six couples worldwide, with male infertility contributing to approximately half of these cases. However, the causes of infertility remain incompletely understood, and current methods of clinical management are cost-restrictive, time-intensive, and have limited success. Artificial intelligence (AI) may help address some of these challenges. In this review, we synthesize recent literature in AI with implications for the clinical management of male infertility.

Recent findings: Artificial intelligence may offer opportunities for proactive, cost-effective, and efficient management of male infertility, specifically in the areas of hypogonadism, semen analysis, and interventions such as assisted reproductive technology. Patients may benefit from the integration of AI into a male infertility specialist's clinical workflow. The ability of AI to integrate large volumes of data into predictive models could help clinicians guide conversations with patients on the value of various treatment options in infertility, but caution must be taken to ensure the quality of care being delivered remains high.

Keywords: Artificial Intelligence; ChatGPT; Machine Learning; Male Infertility.

Fig. 1

Synthesis of uses of AI in male infertility management. Machine learning is a subset of AI that involves assimilating knowledge from large amounts of data to create automated algorithms with predictive capabilities and offers the potential to identify new insights from data that may otherwise be missed. Machine learning can be supervised, in which the outputs (results) are labeled, and classification and regression algorithms are employed to address problems. Specific supervised learning models include decision trees, which are a classification technique (including random forest, gradient boosting, and decision trees including the XGBoost algorithm), AI based image recognition (which can be used in analyzing sperm), regression models, stacking generalization, support vector machines, Naive Bayes, genetic algorithm (an optimization technique inspired by natural selection), Bayesian network, and k nearest neighbors. In unsupervised machine learning, outputs are not labeled, so models can be useful for more descriptive tasks, as they can find relationships in data structures without a measured outcome. Specific unsupervised learning models include the Gaussian mixture model, K means, and principal component analysis. Finally, reinforcement learning involves a computer analyzing available data, deriving rules, and optimizing outcomes; each time the computer receives feedback about its own performance, it can improve subsequent performances via trial and error. Deep learning is a subset of supervised machine learning using an extensive neural computational network simulating the human brain and capable of discovering intricate structures in large datasets. In it, there are layers: the input layer (where information is received), the hidden layer (where processing and pattern extraction takes place), and the output layer (where final network outputs are generated). Deep learning models include artificial neural networks (a broader category of deep learning used for a variety of classification and recognition tasks), convolutional neural networks (which automatically learn spatial hierarchies of features and commonly used for image recognition; UNet is included within this), deep convolutional neural networks (which are similar to convolutional neural networks but with additional layers; including MobileNet), and region based convolutional network (which employ CNN but are suited to analyze temporal and sequential data). Finally, natural language processing leverages the ability of machines to understand, process, and manipulate human language to transcribe audio to text or to extract data from writing, while generative AI can produce content such as text, images, and audio. Examples of NLP are social media and theme analysis, and ChatGPT uses both NLP and is an example of generative AI. Created in BioRender.com