Artificial intelligence in andrology – fact or fiction: essential takeaway for busy clinicians

Abstract

Artificial intelligence (AI) is revolutionizing the current approach to medicine. AI uses machine learning algorithms to predict the success of therapeutic procedures or assist the clinician in the decision-making process. To date, machine learning studies in the andrological field have mainly focused on prostate cancer imaging and management. However, an increasing number of studies are documenting the use of AI to assist clinicians in decision-making and patient management in andrological diseases such as varicocele or sexual dysfunction. Additionally, machine learning applications are being employed to enhance success rates in assisted reproductive techniques (ARTs). This article offers the clinicians as well as the researchers with a brief overview of the current use of AI in andrology, highlighting the current state-of-the-art scientific evidence, the direction in which the research is going, and the strengths and limitations of this approach.

INTRODUCTION

Artificial intelligence (AI) is considered a branch of engineering that attempts to model intelligent behavior to solve complex problems using computers, with minimal intervention of human intelligence. In recent years, research on the use of AI has grown rapidly in all fields, and this innovation has profoundly influenced health care. In medicine, the use of AI can be distinguished into (1) a virtual branch, represented by machine learning, which, through the use of mathematical algorithms, improves learning through experience; and (2) a physical branch, which is represented by robots that help surgeons during surgery or monitoring treatments (Figure 1).1

Figure 1.

Schematic representation of the main areas of application of artificial intelligence’s physical and virtual branches in andrology. MRI: magnetic resonance imaging; ART: assisted reproductive technique.

In principle, there are four different types of machine learning: (1) the supervised learning used when the desired outcome is known; (2) the unsupervised learning used when the target outcome is unknown; (3) the semisupervised learning, particularly used in medical imaging when both labeled and unlabeled data exist; and (4) the reinforcement learning, where the algorithm is trained for a specific task.2

The classic machine learning models used in medicine are linear regression, logistic regression, and the decision tree with its evolution, the random forest. The linear regression model analyzes the correlations between features, and the logistic regression model analyzes the relationship between certain characteristics and the desired outcome. Decision trees and random forests are generally used for classification purposes.2 Other classical machine learning methods include the support vector machine (SVM), which is a regression and classification model that, using kernel functions, transforms a nonseparable problem into a separable problem that is easier to solve.3 On the other hand, dimensionality reduction techniques help transform large set of data with many variables to smaller datasets, which preserve a good deal of the information present in the original source. These include principal component analysis (PCA), uniform manifold approximation and projection (UMAP), and t-distributed stochastic neighbor embedding (t-SNE).3

The artificial neural network (ANN) represents an evolution of classic machine learning, in which biological neural networks inspire the algorithm. The ANN employs nodes, representing neurons, which communicate with each other through specific connections. Each of these connections has a different weight depending on its ability to provide the desired outcome.2 Finally, generative AI, based on AI algorithms that are able to generate new outputs on the basis of the data they have been trained on, also deserves to be mentioned. In this context, generative pretrained models (GPTs), such as the widely known ChatGPT, are the subject of study and controversy in various areas of health.4

In reproductive medicine, AI has been widely used in assisted reproductive technique (ART) to improve the success rate of procedures.5 Moreover, a recent in-depth review conducted by the Global Andrology Forum (GAF) underscored the importance of AI in andrology.6 For example, AI has been applied to generate new diagnostic questionnaires, image classifiers, and disease prediction models in erectile dysfunction (ED).7 This mini review is designed to offer a concise overview of the diverse applications in this field. Our goal is to provide physicians with clear insights into the dynamically evolving and complex realm of AI.

ROLE OF AI IN MALE INFERTILITY DIAGNOSIS

One of the important applications of AI in male infertility diagnostics is its use in the computer-aided sperm analyzers (CASA). These devices are considered useful tools for the rapid analysis of many samples, allowing a reduction in interoperator variability and, therefore, with high reproducibility.6 It also provides a more accurate measure of sperm motility and detailed kinematics. The latter aspect has been highlighted in a recent animal model study that showed how the application of a t-SNE model was able to predict fertility with high accuracy when different sperm kinetic variables detected by the CASA system (curvilinear velocity, straight-line velocity, amplitude of lateral head displacement, and beat-cross frequency) were considered.8

However, although the use of CASA system is promising, they still have limitations that do not allow them to replace the human operators. These include lack of standardization and poor accuracy in assessing sperm morphology and concentration in samples with high viscosity, severe oligozoospermia, or in presence of many round cells, debris, and agglutination.9 In addition, the results can be affected by the type of slide preparation, chamber depths, loading methods, and the representativeness of the evaluated fields.10 This clarifies why the World Health Organization (WHO) 6^th manual in 2021 only recommended CASA systems as advanced tools for examining sperm motility and kinematics.11 However, in semen analysis, AI has proven to be useful in the prediction of male infertility. In detail, several machine learning models have been tested for this purpose, showing that random forest achieves an optimal accuracy and area under the curve (AUC) of 90.47% and 99.98%, respectively, in fertility prediction.12 Furthermore, AI has been used in the evaluation of lifestyle and environmental factors that predict alterations of human fertility, demonstrating, for example, that characteristics such as age, alcohol consumption, cigarette smoking, and sedentary lifestyle are more effective at predicting an alteration in future fertility than having had childhood illnesses.13

Finally, machine learning algorithms and deep learning methods have been developed to predict sperm with a high DNA fragmentation rate, a known cause of male infertility and ART failure.14

AI has also been used in flow cytometry. The latter is used in male reproduction to evaluate biofunctional sperm parameters including DNA fragmentation, mitochondrial membrane potential, oxidative stress, and membrane peroxidation.15 To date, there are software programs such as FlowJo™ or Cytobank™ that include machine learning tools such as concatenation, t-SNE, and clustering that allow the analysis of flow cytometry data at a single-cell level to offer useful information on the biology of sperm function.16

USE OF AI IN ARTS

Several classical machine learning and deep learning approaches have been used to analyze embryo quality and assist in selecting embryos to be transferred to the uterus.17 Additionally, AI has been used to predict the success of the procedure.17 For example, a study, through the creation of an ANN trained with 12 features, including the woman’s age, the total dose of gonadotropin administered, endometrial thickness, and the number of top-quality embryos, highlighted a cumulative sensitivity and specificity of 76.7% and 73.4%, respectively, in predicting live births.18 Furthermore, the role of AI in selecting spermatozoa to be used in ART appears promising. Indeed, it has been demonstrated that deep learning or SVM methods have high sensitivity and specificity in selecting spermatozoa with high-quality morphology.19 AI is also useful in assessing sperm motility with CASA systems.9 Therefore, the use of an integrated approach mediated by AI, which takes into account all these parameters simultaneously, could constitute an important tool to embryologists for the selection of the best spermatozoa for ART. This in turn would result in an improved pregnancy rate and live birth rate.19

USE OF AI IN ANDROLOGICAL DECISION-MAKING

AI has been used in andrology to assist in clinical decision-making. A recent study on 240 patients who underwent varicocele repair identified a random forest model with high accuracy for predicting patients who would improve their sperm parameters after the intervention.20 The study showed that serum follicle-stimulating hormone (FSH) levels and the presence of bilateral varicocele are two fundamental parameters for the predictive capacity of the AI-derived model. Furthermore, the model was able to predict upgrades in 87% of patients deemed likely to improve following varicocele repair. This approach promises to help select patients for surgery, thus reducing the rate of overtreatment.

Another study considered patient weight, age, surface area volume, and FSH levels as variables in the decision-making process to identify patients with nonobstructive azoospermia (NOA) who may or may not benefit from a sperm extraction attempt. The gradient-boosted trees machine learning approach demonstrated superior accuracy when compared to a multivariate logistic regression model.21

Similarly, AI algorithms have been used to improve the standardization of the Gleason score in patients with prostate cancer. In particular, a study used 698 prostate biopsies to train an AI algorithm based on convolutional neural networks (CNN) and 37 biopsies to test its validity. The results demonstrated that the algorithm had high precision in detecting tumor areas and correctly assigning Gleason patterns, with similar accuracy to that of two pathologists. This approach could eliminate the interoperator variability that is inevitably present when the diagnosis is made with the human eye.22

The diagnostic capacity of AI has also been studied in ED. In a study involving 2832 patients with ED and 2832 without ED, the authors designed a clinical decision support system (CDSS) based on integrated genetic algorithm and SVM, which showed high sensitivity, specificity, and, consequently, accuracy in predicting the incidence of ED.23 In detail, the model proposed by the authors using salient features such as age, presence of comorbidities, and other comorbidity-related variables (age of diagnosis, follow-up duration, and frequency of physician visits of the comorbidities) was able to predict the incidence of ED. In addition to assisting with diagnosis and the decision-making process, AI has demonstrated its capacity to contribute to developing new questionnaires and evaluating ED through medical imaging techniques.7 Indeed, a study evaluating the accuracy of different ED questionnaires using three different machine learning methods (Naive Bayes, k-Nearest Neighbors, and SVM) found that a visual scale questionnaire had superior accuracy in assessing the severity of ED compared with more complex questionnaires such as the International Index of Erectile Function-5 (IIEF-5) and the Massachusetts Male Aging Study (MMAS) Sexual Activity Questionnaire.24

USE OF AI IN ANDROLOGICAL IMAGING

The use of AI in diagnostic imaging in andrology is also noteworthy. Extensive use of machine learning models has been made to study prostate cancer using magnetic resonance imaging (MRI), and AI has been shown to be useful in the process of segmentation, lesion detection, and aggressiveness prediction.25 For example, it was observed that machine learning models, using quantitative imaging parameters such as perfusion maps, apparent diffusion coefficient, and absolute T2 signal intensities, can predict the presence of clinically significant cancer, defined as Gleason score ≥3+4, better than Prostate Imaging Reporting and Data System.26

AI has been employed in andrological imaging in fields other than oncology. In a recent study, AI confirmed that the testicular echo structure reflected the patient’s reproductive function, demonstrating that a radiomics approach to testicular echotexture could predict testicular spermatogenic capacity and correlate with pituitary function,27 highlighting the importance of ultrasound diagnostics in studying male fertility.

Another study using the SVM model demonstrated that diffusion tensor imaging indices of some brain areas evaluated by MRI could prove useful in diagnosing ED caused by an alteration of the veno-occlusive mechanisms.28

USE OF AI IN ANDROLOGICAL SURGERY

The physical branch of AI, notably using robotics in andrological surgery, deserves a mention. In this context, augmented reality techniques of prostate reconstruction based on multiparametric MRI images were tested. The association between robotics and these techniques promises to be a useful approach capable of improving the outcome of patients undergoing prostatectomy, tailoring the procedure to the patient.29 Robotic surgery also has a role in infertility treatment, where it could be useful for various microsurgical procedures such as vasectomy reversal, varicocelectomy, testicular sperm extraction, and spermatic cord denervation. The advantages of robotics in this field are the elimination of operator tremors, three-dimensional visualization, and the reduced need for qualified surgical assistance.30 However, some limitations have prevented the rapid integration and diffusion of robots in surgery. Among the main limits are the lack of studies with a substantial number of cases demonstrating the superiority of using robots in this type of surgery, skepticism from experienced surgeons in using a robot for delicate tissue manipulation, and the high costs of robotic surgery compared to traditional surgery.30 The current main fields of application of AI in andrology are summarized in Figure 1.

STRENGTHS AND LIMITATIONS OF THE USE OF AI

The advantages of AI are numerous, including greater precision and accuracy of the diagnosis, making it possible to resolve even those cases in which there is no agreement between physicians. Its use can also help increase a test’s diagnostic sensitivity and specificity and assist the physician in disease management decision-making.31 This can reduce medical errors and, consequently, medical costs, morbidity, and mortality.32 Its deployment in rural areas could also bridge the gap between rural and urban health care, overcoming the lack of expert physicians in these areas and thus improving the quality of care.33

However, to date, the use of AI in medicine has several limitations, such as the high cost of the hardware that manages the AI algorithms34 or the robots used in surgery.30 Additionally, concerns may arise about the safety and reproducibility of the software and ethical-legal problems related to protecting patients’ privacy.34 Furthermore, the risk cannot be ruled out that there may be a future reduction in clinicians’ skills who may rely too much on AI for the clinical decision-making process.34 Finally, it should be considered that, to date, these algorithms have often been tested on limited populations, so the possibility of translating their effectiveness to the general population is still questionable, thus reducing the validity and reliability of this approach.35

FUTURE PERSPECTIVES ON AI

In the field of reproduction, AI could enable increasingly precise execution of ART, starting from a personalization of the stimulation protocol to a more accurate selection of gametes to be used during the procedure, up to the identification of the best embryo to transfer, all to improve pregnancy and live birth rates.36 Furthermore, in the context of NOA, the use of AI could be useful in identifying, through the use of biomarkers such as leptin and FSH, patients who could benefit from sperm retrieval procedures37 and helping to identify azoospermic patients who require further genetic testing.38 Finally, developing increasingly objective and specific predictive algorithms could become a useful tool for clinicians to assist them in the clinical decision-making process, thus reducing the cost of care and allowing patient-tailored medicine to improve overall health.39

CONCLUSIONS

The evidence on AI that has emerged in recent years suggests a role for this technology in supporting physicians in managing some andrological problems. To date, the main areas of application of AI in andrology include its use in improving the success rate of ART and supporting the decision-making process among physicians. Another important field of application is the integration of machine learning algorithms into imaging to improve its diagnostic performance. The inevitable progress of technology will probably make it possible to respond to many issues that clinicians currently face when managing patients in the future. Just thinking of the current low success rate of ART, one could see a rapid increase in the future through machine-learning approaches. Despite the increasing number of studies highlighting the role of AI in managing andrological diseases, many studies are still needed to validate its effectiveness. The main limitation of all machine learning studies is the small sample size of selected patient groups, which precludes the ability to apply these algorithms to the general population. Further studies to validate machine learning models on large populations are required to confirm the reliability and validity of this approach. The creation of properly designed multicenter studies could probably overcome this issue. The physicians’ task in the future will be to work in an integrated manner with machine learning without blindly relying on it. Otherwise, the clinicians who passively accept relying on AI without integrating it with their professional background would inevitably face professional disqualification with a consequent loss of trust and empathy, which are fundamental prerequisites for every physician, especially andrologists.

AUTHOR CONTRIBUTIONS

AA and AEC conceptualized the manuscript, were responsible for revising and editing the manuscript, and supervised the drafting of the article. AC was responsible for research methodology, data curation, and writing of the original draft. RC contributed to data curation and manuscript review and editing. R Saleh and R Shah validated the study and contributed to the review and editing of the manuscript. All authors read and approved the final manuscript.

COMPETING INTERESTS

All authors declare no competing interests.