Integrating artificial intelligence in surgical sperm retrieval techniques: A narrative review

Abstract

Nonobstructive azoospermia (NOA) is a serious form of male infertility with therapeutic options limited to trials of endocrine manipulations and repertoire of surgical interventions, also known as surgical sperm retrieval (SSR) procedures. Despite its invasive nature, SSR remains crucial in the management of NOA, offering infertile males the opportunity of fathering their biological children using assisted reproductive technologies. Success rates of SSR are variably governed by several factors including the genetic background, preoperative endocrine optimization, testicular histopathology, surgeon's microsurgical expertise, and laboratory technological and technical team's capability. This paper explores the significant role of artificial intelligence (AI) in the process of sperm retrieval among NOA patients. The role of AI has evolved from basic predictive models used for outcome assessment and patient counseling, to advanced image processing capabilities for assessing sperm parameters, and now to cutting-edge applications in identifying the rare sperm present in the azoospermic microdissection testicular sperm extraction tissue samples.

Graphical abstract

Keywords

artificial intelligence / micro-TESE / nonobstructive azoospermia / sperm retrieval

1 INTRODUCTION

Male infertility is a serious medical condition affecting males worldwide and is on the rise; with nearly 1 in 20 men suffering from reduced fertility[1,2]. The clinical profile varies according to the severity of the condition, with nonobstructive azoospermia (NOA) considered the most serious form of male infertility[3]. Defined as the absence of sperm in the ejaculate due to failure of production, NOA occurs in approximately 1% of the general population and 15% of patients with infertility[4]. Despite its poor prognosis, NOA could still be managed with variable degrees of success. With the advent of artificial reproductive techniques (ART), many infertile men are given the chance to father their biological children, and as the number of sperm required for ART is minimal, introducing surgical sperm retrieval (SSR) procedures in patients with NOA is crucial. Once sperm are retrieved, they can be used in ART[5]. A growing number of SSR techniques can be offered to NOA patients, with microdissection testicular sperm extraction (micro-TESE) being the most up-to-date and the gold standard approach[5]. Despite its significant role in management, the outcome of micro-TESE is subject to multiple factors that are not only limited to the surgeon's expertise, but also the optimal laboratory techniques and environment[6,7].

Recently, the healthcare industry has been grappling with the adoption of artificial intelligence (AI), which has disrupted contemporary healthcare. AI refers to the process of developing computer systems that are capable of efficiently performing complex tasks that usually require human intelligence and dexterity[8] and is defined as the capability of the engineered systems to acquire, process and apply knowledge and skills according to the International Organization for Standardization/International Electrotechnical Commission technical report[9]. AI has revolutionized NOA management, offering a wide range of innovative solutions, from numerous sophisticated predictive models to guide physicians in counseling their patients, to the implementing of ground-breaking laboratory innovations that utilize complex computational algorithmic models to accurately assess semen samples, eliminating human error and subjectivity[10]. The most up-to-date and innovative measures include image recognition technology, which is used to analyze micro-TESE tissue samples, significantly improving the capability of retrieving the rare sperm within a sample[11]. However, there remains a significant challenge in identifying rare sperm in the complex micro-TESE testicular specimens, with an existent possibility of false-negative results, which could result from human error or limited laboratory capabilities[7]. By leveraging AI, healthcare providers can potentially transform the landscape of male infertility treatment, offering improved outcomes and personalized care for patients suffering from this complex condition[12,13]. This will potentially revolutionize this aspect of management, yet its universal applicability and affordability remain uncertain.

2 CONVENTIONAL SSR PROCEDURES

SSR is defined as the group of surgical procedures that are used to retrieve sperm from the testis or the epididymis[14]. This approach is generally indicated where no viable sperm are present in the semen and hence, commonly offered in the context of ART when sperm are absent in the ejaculate, either due to an obstructive or nonobstructive pathology[15]. Moreover, SSR may also be considered in patients with an adequate number of ejaculated sperm, yet intratesticular sperm, which commonly exhibit better DNA quality is preferred especially in the context of elevated sperm DNA fragmentation and severe seminal parameters abnormalities[16].

In cases of obstruction, such as obstructive azoospermia (OA), for example, after vasectomy, SSR is more commonly indicated and rates of retrieval may reach up to 100%, as obstructed testes can remain functional and produce enough sperm to make SSR procedures less challenging[17]. In the clinical context of testicular failure, that is, NOA, the testes fail to produce sperm, commonly due to genetic aberrations, for example, Klinefelter syndrome, Y-chromosome microdeletion and translocations[18]. Acquired causes can also contribute to the development of NOA, including undescended testis, mumps orchitis and chemotherapy-induced testicular failure[19].

Considering the different nature, NOA is associated with the considerably less favorable outcomes when it comes to sperm retrieval, compared to OA, primarily because of the underlying testicular failure where sperm production is either severely compromised or totally absent[17,20]. One of the most performed SSR procedures is fine needle aspiration-testicular sperm aspiration (FNA-TESA), which is a less invasive and less technically demanding procedure, entailing percutaneous tissue aspiration using an 18 gauge needle to retrieve testicular tissues and subsequent examination for the presence or absence of sperm[21] (Figure 1). FNA-TESA is commonly indicated in OA or in those who have sperm in their ejaculate, while aiming to retrieve better-quality sperm for ART. Conversely, FNA-TESA is less indicated in NOA, except in the context of testicular mapping, which is performed to outline active sperm-harboring regions within the testis[22].

Another conventional SSR technique is conventional testicular sperm extraction (cTESE), which is the surgical excision of testicular specimens under nonmagnified visual control[23] (Figure 2).

To improve SSR rates in the context of NOA, Schlegel introduced an advanced and widely adopted technique known as micro-TESE, which utilizes the magnifying power of the surgical microscope[6] (Figure 3). Micro-TESE entails an extensive surgical exploration of the testes, exposing the seminiferous tubules after bivalving the testes and undergoing a series of delicate microdissections under the guidance of the surgical microscope, aiming for morphologically distinct seminiferous tubules, which may appear opaquer and more dilated seen under high-power magnification, suggesting the presence of active spermatogenesis[5] (Figure 4).

Following the surgical microdissection, the tissues are gently handled to the ART lab team for examination. The laboratory procedures used in handling the seminiferous tubules involve a complex process of either mechanical tissue maceration, or enzymatic processing using collagenase and centrifugation that would result in releasing the testicular sperm into the field for proper visualization and isolation[24–26].

Over the years, micro-TESE has become the gold standard SSR technique for patients with NOA despite its inherent technical challenges, requiring advanced microsurgical expertise and a steep learning curve[27]. In a systematic review, SSR ranged from 42.9% to 63% for microsurgical retrievals compared to 16.7%–45% in conventional biopsies[28], with variable degrees of SSR success in Sertoli cell only syndrome (22.5%–41%), early maturation arrest (27%–40%), late maturation arrest (27%–86%), and hypospermatogenesis (73%–100%)[5,28–32]. In a systematic review and meta-analysis by Bernie et al., pooled data from 1890 patients included in 15 studies, it was shown that micro-TESE resulted in 1.5 times higher sperm retrieval rate compared to cTESE, while TESA was two times less likely to retrieve sperm compared to cTESE[32]. SSR success rates can be influenced by several factors, including the necessity of the optimal laboratory techniques, advanced technology and skilled staff, which can yield a false-negative result if not properly controlled, potentially skewing the outcome[7].

3 AI IN REPRODUCTIVE MEDICINE

Recently, the healthcare industry has witnessed an unprecedented disruption due to the rapid adoption of AI, which has up-scaled progressively to involve diverse contemporary medical domains[33]. Being an innovative tool, AI is a comprehensive term referring to the computer utility in performing tasks that generally require reasoning and learning[34]. Such innovative applications are pioneered to assist physicians in various healthcare settings and to deliver a tailored and personalized patient-centered care[35]. AI is regarded as a machine learning (ML) process, which after a series of training sessions, can acquire learning and problem-solving capabilities. Previously, available computational models were confined to a predictive functionality, helping healthcare professionals identify undetectable patterns to facilitate diagnosis and management[36,37]. Subsequently, more complex computational algorithms were introduced and developed advancing ML with more sophisticated and profound capabilities, which facilitated diagnosis and management[38].

In essence, ML is conceptualized around the software algorithms performing training sessions where they analyze statistical outcomes from labeled test sets and compare against a specific outcome to identify a correlation or a discernible pattern[39]. ML can be categorized as supervised when the used training data is compared against a specified outcome. Conversely, it is unsupervised when no labeling data or human interference exist, and the model is left to recognize patterns[40]. ML is not confined to learning, recognizing patterns or problem solving, but also being able to automatically undergo improvements[41].

In the field of andrology, supervised ML models have been developed to assist in the diagnosis and management of male infertility. Krenz et al. developed a prediction model used on azoospermic patients to identify patients with Klinefelter syndrome[42], while Ory et al. developed a random forest model that can predict improvement in semen parameters following varicocele repair[43]. Furthermore, AI has become increasingly engaged in various pathological realms, leveraging the “-omics” data to explore more complex biological realms. With that being said, the innovative cloud-based technology and high-speed data transfer have rendered the application and utility of AI models more seamless and accessible with significant feasibility of monitoring and improvements[35,44].

For instance, in the field of radiomics, where medical images are more profoundly processed and analyzed using complex algorithms, a more precise characterization of the image is obtained, but can help in patient stratification according to the different risk groups[45]. Deep learning (DL) as a subtype of AI uses multiple layers of data, for example, neural networks, which can analyze and identify intricate patterns existing in complex and detailed data models[46]. Neural networks are used for different forms of data, for instance, in a type of DL model, convolutional neural networks (CNNs) are used to analyze spatial data arranged in a grid pattern, rendering them particularly effective for image processing[47], while in recurrent neural networks, where inputs are memorized, sequential data are used instead (e.g., time)[48].

4 AI PREDICTIVE MODELS IN REPRODUCTIVE MEDICINE

In the field of reproductive medicine, several factors influence the outcome, that is, healthy live births. Male factor contributes to around 30%–40% of the couple's overall infertility, with female factors contributing at nearly similar rates[49]. Currently, semen parameters are commonly assessed using the conventional semen analysis (SA), which is an unreliable tool, incapable of discerning between fertile and infertile males and subject to significant variability that warrants a need for repeated testing[50]. Sperm DNA and chromatin structural studies have emerged as a supporting tool in assessing the fertilizing capacity of the human sperm[51]. Genetic and epigenetic background can significantly be involved in the different sperm abnormalities including sperm maturation, acrosome reaction, and fertilization[52–54].

Furthermore, normal spermatogenesis is highly dependent on normal endocrine function, essentially follicle stimulating hormone and testosterone[55]. Consequently, multiple factors interact to influence male fertility, and addressing each factor individually is insufficient to fully understand male fertility potential. In this context, AI predictive models provide a valuable tool by analyzing existing data to predict and anticipate outcomes more effectively. Moreover, numerous clinical decision support algorithms have been offered to assist in diagnosis and management. Artificial neural networks (ANNs) can be sought in evaluating in vitro fertilization (IVF) data, aiming to predict treatment success, as they are highly effective in handling complex data with high levels of variability[56].

In a study by Vogiatzi et al., 12 statistically significant parameters were collected from 426 IVF/intracytoplasmic sperm injection cycles and used in the construction of ANN that resulted in a 76.7% sensitivity and 73.4% specificity, which was further cross validated to establish a sensitivity and specificity of 69.2% ± 2.36% and 69.19% ± 2.8%, respectively[57]. In a study by Yi et al., sperm morphology was classified using ANNs as either normal or abnormal[58]. In another study by Wang et al., images of sperm in a training set were used in a linear and nonlinear regression models to predict DNA fragmentation from sperm morphology (r = 0.558, and 0.62, respectively)[59]. Kandel et al. used computational methods to predict IVF pregnancy outcomes according to sperm ultrastructure[60].

In the context of NOA, an ANN model was developed to predict the odds of sperm retrieval before micro-TESE, utilizing data from 1026 patients performing micro-TESE. This model was capable of predicting the outcome in approximately 60% of cases, with a receiver operating characteristic (ROC) area of 0.641. Furthermore, a nomogram was produced for such purpose with an AUC of 0.59[61]. These values, though promising for future applications, yet require further refinement. Table 1 summarizes key AI-based algorithms that enhance the role of ML across various medical fields.

5 AI-ENHANCED SPERM IMAGE RECOGNITION

In the context of sperm detection, previous algorithms dedicated to sperm detection used several image processing techniques, including, for example, using the combined Lambertian optical model of light reflection and a mathematical morphologic image processing framework used in segmenting sperm cell images[64], or using an adaptive local threshold and ellipse detection, which entails applying locally adjusting thresholds to manage uneven illumination in a specimen and identifying objects with elliptical shape (i.e., sperm head) in the thresholded images[65]. Unfortunately, the intensive computational demands for image processing, eventually limit the number of samples that can be processed in a given time (throughput)[66].

With advancements in predictive models, AI has evolved beyond numerical data-based processing and analysis to image recognition, achieving high precision and accuracy in detecting intricate structures and shapes. Image processing is challenged by the capability of attaining powerful resolution, a fact that requires large data training sets, and powerful computational capabilities to minimize the blurriness of the image. For this purpose, Dobrovolny et al. proposed a supervised generative adversarial network system, which can generate high-resolution images from low-resolution input, without interfering with the pathological invariance[67]. Furthermore, in an attempt to reduce the subjectivity of conventional SA results, a study by Dobrovolny et al., developed a deep neural network algorithm using a dataset comprised of World Health Organization SA data, live sperm video captures, spermatic fatty acid profile, serum phospholipid fatty acid composition and demographic characteristics, from 85 distinct male subjects, to detect healthy sperm, which resulted in a highly precise and rapid model with a mean average precision of 72.15[68].

Sperm motility is considered pivotal in the assessment of sperm vitality and function, yet in the most advanced measuring systems, the multisperm tracking is lacking, possibly due to the large sperm count, challenging the ability to identify the sperm path along the field[69,70]. This was addressed by numerous projects, for instance, Su et al. used a lens-free on-chip device to undergo sperm tracking in a three-dimensional fashion and was capable in identifying trajectory patterns while imaging appropriately 34 000 sperm trajectories at a rate of 90–140 frames per second[69].

6 TECHNOLOGICAL INNOVATIONS IN AI FOR SPERM IDENTIFICATION DURING SSR

Searching for and identifying sperm in azoospermic samples is a critical and challenging process, as the outcome is paramount in the decision-making process of whether the patient should or should not move forward with sperm retrieval surgery. Ensuring the identification of existing sperm is prudent to prevent undergoing unnecessary surgical retrieval[7,13]. Moreover, sample processing in the laboratory following the delivery of the specimen may take up to 4 h–6 h[71]. The extended lapse of time during sample processing and examination holds a significant risk on the sperm viability[72]. With that being said, the surgical retrieval itself is not the sole determining factor for success in identifying sperm, in fact, laboratory-related factors have a profound impact on sperm retrieval[73]. In one study, sperm were identified in the laboratory in 7% of samples where no sperm had been identified intraoperatively[74].

To address these challenges, Goss et al. trained an AI model to identify sperm within complex tissue sample. They demonstrated that sperm was detected significantly faster per field of examination using the AI model compared to manual identification by trained embryologists (0.019 ± 1.4 × 10⁻⁴ s vs. 22.87 ± 0.98 s, p < 0.0001), with a significant difference in accuracy favoring the AI model over the trained embryologist (89.88 ± 1.56% vs. 83.22 ± 2.02%, p = 0.017)[12].

As previously mentioned, the process of tissue analysis and sperm identification in surgically retrieved azoospermic samples is extremely challenging and is governed by several factors that dictate the success of sperm recovery. False-negative results are common due to the presence of crowded collateral cell density, lack of experience or simply human fatigue. Such an error could have a devastating outset on the patient and the couple[7,74]. Wu et al. developed a computer-aided sperm analysis system, which is based on a DL model, trained to identify sperm in testicular tissue specimens. The model used data generated from images from 702 testicular specimens across 30 patients divided into training, validation, and testing sets (80%, 10%, and 10%, respectively). Compared with embryologists' performance, the model achieved a mean average precision of 0.741 and an average recall of 0.376[75].

Recently, a two-phase proof of concept study was published showcasing a CNN model that was trained using approximately 10 000 sperm images from 8 azoospermic patients to detect sperm within samples. Furthermore, a side-by-side comparison between two azoospermic sample cohorts (n = 4), examined by the AI versus an embryologist and another side-by-side testing by embryologist with and without AI support (n = 4) was performed. Results from this study demonstrated in the first stage that compared to the embryologist, the time taken by AI to detect the sperm in the sample was 0.02 ± 0.30 × 10^–5 s versus 36.10 ± 1.18 s (p < 0.0001) with significantly improved recall of 91.95 ± 0.81% versus 86.52 ± 1.34% (p < 0.001). Results further showed that out of 2660 sperm, the AI identified, in less than 1000th of the time, 1997 sperm compared to only 1937 sperm that were identified by the embryologist.

The way the software operates is by creating a bounding box around the sperm that appears while panning through the slide, reflecting a real-time identification of the sperm (Figure 5). Furthermore, time taken per droplet in the searching process was significantly less for the AI-assisted embryologist (98.90 ± 3.19 s vs. 168.7 ± 7.84 s, p < 0.0001) with 1396 sperm being found compared to 1274 sperm found without AI assistance[13].

The robustness of a model is further strengthened by expanding the capability of detection in unconventional conditions, by adjusting for image augmentation, which involves subjecting the same training-set images to various circumstances, for example, rotation, color variation, image flipping, or changes in focus so as to train the model to identify the desired object even if the circumstances around are changed[75,76]. Furthermore, another team of researchers have successfully developed an ML model that is trained using bright-field microscopy to identify rare sperm in both, semen and micro-TESE tissue samples. This model was developed using a CNN algorithm trained and validated using 35 761 and 7663 image patches, respectively. The model was then tested with 7663 and 7985 image patches, containing numerous and rare sperm samples, respectively.

Moreover, for sperm-only samples, the algorithm achieved a positive predictive value, sensitivity and F1-score at ×10 magnification of 91%, 95.8% and 93.3%, respectively. However, for the sperm-abundant micro-TESE samples, it achieved 84%, 72.2%, and 77.9%, respectively, and for sperm-rare micro-TESE samples, it achieved 84.4%, 86.1%, and 85.2%, respectively[11].

7 CHALLENGES OF AI IN SSR

Despite the innovative measures offered by AI in the realm of SSR, there remain some challenges that would be faced by fertility specialists, urologists, and researchers. This hurdle resides in the specimen variability, which is highly complex requiring extensive training of the model to adapt to the different specimen composition and complex cellular environment[11–13].

8 ETHICAL CONSIDERATIONS

Comprehending the magnitude of the ethical landscape towards the rising application of AI in reproductive medicine is paramount. Informed consents about the whereabouts of the therapeutic plan should clearly be outlined, with adequate transparency allowing for patient contribution in the decision-making process. The significance of the matter resides in the involvement of the offsprings, which is produced following the selection process of an AI-based model. In such a context, with the lack of studies to measure future outcomes, creates a challenge toward evaluating the impact of AI sperm selection, on the potential offspring's health[38].

Furthermore, as numerous factors interplay in the management process of patients when AI is involved, knowing upon which the reasonability should fall is crucial. However, concern regarding the universal adoption of this technology is based on the essence of individual variability[77]. Additionally, addressing the associated financial burden to this expensive utility is paramount to ensure universal applicability across the medical landscape[78].

9 FUTURE DIRECTIONS

AI has shown promising innovative measures in managing male infertility, especially in improving SSR for NOA patients. It is expected that as these AI-driven techniques advance further, their role in increasing the SSR precision will subsequently evolve with further improvements in the SSR-related precision by refining algorithms to further adapt with the complexity and diversity of testicular specimens. This refinement is expected to reduce both false positives and false negatives, ultimately leading to more accurate results. The expected improvement in AI-based models will possibly include expanding and diversifying datasets to better train models, making them become integral in the decision-support process[63]. This may involve integrating sperm ultrastructural data with functional information, such as fertilization capabilities, to provide predictive outcomes based on these parameters, thereby indicating which sperm are most likely to yield the best results in ART[10,79] (Table 2).

10 CONCLUSION

In conclusion, NOA remains a challenging medical disorder, not only requiring an experienced medical team, but also requires an efficient and technologically advanced laboratory. The capability of identifying rare sperm within the complex testicular specimens is paramount in enabling azoospermic patients to father their biological offspring. AI has emerged as a disruptive force transforming contemporary medical practices, evolving as a revolutionary tool using its computational algorithms to perform robust image processing and object identification. This functionality has recently been implemented in identifying rare sperm in testicular micro-TESE suspensions, increasing the rate of sperm recovery and hence, the chances of males previously rendered infertile to achieve biological parenthood.

References

[1] Ravitsky V, Kimmins S. The forgotten men: rising rates of male infertility urgently require new approaches for its prevention, diagnosis and treatment. Biol Reprod. 2019;101(5):872–874.

[2] Levine H, Jørgensen N, Martino-Andrade A, Mendiola J, Weksler-Derri D, Mindlis I, et al. Temporal trends in sperm count: a systematic review and meta-regression analysis. Hum Reprod Update. 2017;23(6):646–659.

[3] Caroppo E, Colpi GM. Male infertility: a review of key papers appearing in the reproductive medicine and andrology section of the journal of clinical medicine. J Clin Med. 2023;12(6):2366.

[4] Sharma M, Leslie S. Azoospermia. StatPearls. Treasure Island: StatPearls Publishing; 2024.

[5] Flannigan R, Bach PV, Schlegel PN. Microdissection testicular sperm extraction. Transl Androl Urol. 2017;6(4):745–752.

[6] Schlegel PN. Testicular sperm extraction: microdissection improves sperm yield with minimal tissue excision. Hum Reprod. 1999;14(1):131–135.

[7] Samuel R, Badamjav O, Murphy KE, Patel DP, Son J, Gale BK, et al. Microfluidics: the future of microdissection TESE? Syst Biol Reprod Med. 2016;62(3):161–170.

[8] Hamet P, Tremblay J. Artificial intelligence in medicine. Metabolism. 2017;69S:S36–S40.

[9] ISO/IEC. ISO/IEC TR 24028:2020. Accessed 25 Oct 2024.

[10] Panner Selvam MK, Moharana AK, Baskaran S, Finelli R, Hudnall MC, Sikka SC. Current updates on involvement of artificial intelligence and machine learning in semen analysis. Medicina. 2024;60(2):279.

[11] Lee R, Witherspoon L, Robinson M, Lee JH, Duffy SP, Flannigan R, et al. Automated rare sperm identification from low-magnification microscopy images of dissociated microsurgical testicular sperm extraction samples using deep learning. Fertil Steril. 2022;118(1):90–99.

[12] Goss D, Vasilescu S, Vasilescu P, Sacks G, Gardner D, Warkiani M. O-136 artificial intelligence to assist in surgical sperm detection and isolation. Hum Reprod. 2023;38(Suppl_1):dead093.163.

[13] Goss DM, Vasilescu SA, Vasilescu PA, Cooke S, Kim SH, Sacks GP, et al. Evaluation of an artificial intelligence-facilitated sperm detection tool in azoospermic samples for use in ICSI. Reprod Biomed Online. 2024;49(1):103910.

[14] Shah R. Surgical sperm retrieval: techniques and their indications. Indian J Urol. 2011;27(1):102–109.

[15] Kresch E, Efimenko I, Gonzalez D, Rizk PJ, Ramasamy R. Novel methods to enhance surgical sperm retrieval: a systematic review. Arab J Urol. 2021;19(3):227–237.

[16] Esteves SC, Majzoub A, Agarwal A. More good than harm should be expected when Testi-ICSI is applied to oligozoospermic men with post-testicular sperm DNA fragmentation. Transl Androl Urol. 2017;6(Suppl 4):S381–S384.

[17] Coward RM, Mills JN. A step-by-step guide to office-based sperm retrieval for obstructive azoospermia. Transl Androl Urol. 2017;6(4):730–744.

[18] Peña VN, Kohn TP, Herati AS. Genetic mutations contributing to non-obstructive azoospermia. Best Pract Res Clin Endocrinol Metab. 2020;34(6):101479.

[19] Cocuzza M, Alvarenga C, Pagani R. The epidemiology and etiology of azoospermia. Clinics. 2013;68 Suppl 1(Suppl 1):15–26.

[20] Ishikawa T. Surgical recovery of sperm in non-obstructive azoospermia. Asian J Androl. 2012;14(1):109–115.

[21] Cito G, Coccia ME, Sessa F, Cocci A, Verrienti P, Picone R, et al. Testicular fine-needle aspiration for sperm retrieval in azoospermia: a small step toward the technical standardization. World J Men's Health. 2019;37(1):55–67.

[22] Beliveau ME, Turek PJ. The value of testicular “mapping” in men with non-obstructive azoospermia. Asian J Androl. 2011;13(2):225–230.

[23] Esteves SC. Microdissection TESE versus conventional TESE for men with nonobstructive azoospermia undergoing sperm retrieval. Int Braz J Urol. 2022;48(3):569–578.

[24] World Health Organization. WHO laboratory manual for the examination and processing of human semen. 6th ed. Geneva: World Health Organization; 2021.

[25] Modarresi T, Sabbaghian M, Shahverdi A, Hosseinifar H, Akhlaghi AA, Sadighi Gilani MA. Enzymatic digestion improves testicular sperm retrieval in non-obstructive azoospermic patients. Iran J Reprod Med. 2013;11(6):447–452.

[26] Vloeberghs V, De Munck N, Racca A, Mateizel I, Wouters K, Tournaye H. Enzymatic tissue processing after testicular biopsy in non-obstructive azoospermia enhances sperm retrieval. Human Reproduction Open. 2023;2023(4):hoad039.

[27] Fraietta R. Microdissection is the best way to perform sperm retrieval in men with non-obstructive azoospermy? | Opinion: yes. Int Braz J Urol. 2018;44(6):1063–1066.

[28] Deruyver Y, Vanderschueren D, Van der Aa F. Outcome of microdissection TESE compared with conventional TESE in non-obstructive azoospermia: a systematic review. Andrology. 2014;2(1):20–24.

[29] Bernie AM, Shah K, Halpern JA, Scovell J, Ramasamy R, Robinson B, et al. Outcomes of microdissection testicular sperm extraction in men with nonobstructive azoospermia due to maturation arrest. Fertil Steril. 2015;104(3):569–573.

[30] Caroppo E, Colpi EM, Gazzano G, Vaccalluzzo L, Scroppo FI, D'Amato G, et al. Testicular histology may predict the successful sperm retrieval in patients with non-obstructive azoospermia undergoing conventional TESE: a diagnostic accuracy study. J Assist Reprod Genet. 2017;34(1):149–154.

[31] Kavoussi PK, Hunn C, Gilkey MS, Chen SH, Kavoussi KM, Wininger JD, et al. Sertoli cell only syndrome induced by a varicocele. Transl Androl Urol. 2019;8(4):405–408.

[32] Bernie AM, Mata DA, Ramasamy R, Schlegel PN. Comparison of microdissection testicular sperm extraction, conventional testicular sperm extraction, and testicular sperm aspiration for nonobstructive azoospermia: a systematic review and meta-analysis. Fertil Steril. 2015;104(5):1099–1103.e3.

[33] Poalelungi DG, Musat CL, Fulga A, Neagu M, Neagu AI, Piraianu AI, et al. Advancing patient care: how artificial intelligence is transforming healthcare. J Pers Med. 2023;13(8):1214.

[34] Thomason R. Logic-based artificial intelligence. In: Zalta EN, Nodelman U, editors. The Stanford encyclopedia of philosophy. 2024th ed. Stanford: Metaphysics Research Lab, Stanford University; 2024.

[35] Ghayda RA, Cannarella R, Calogero AE, Shah R, Rambhatla A, Zohdy W, et al. Artificial intelligence in andrology: from semen analysis to image diagnostics. World J Men's Health. 2024;42(1):39–61.

[36] Kaplan A, Haenlein M. Siri, Siri, in my hand: Who's the fairest in the land? On the interpretations, illustrations, and implications of artificial intelligence. Bus Horiz. 2019;62(1):15–25.

[37] Sidey-Gibbons JAM, Sidey-Gibbons CJ. Machine learning in medicine: a practical introduction. BMC Med Res Methodol. 2019;19(1):64.

[38] Rolfes V, Bittner U, Gerhards H, Krüssel JS, Fehm T, Ranisch R, et al. Artificial intelligence in reproductive medicine—an ethical perspective. Geburtshilfe Frauenheilkd. 2023;83(1):106–115.

[39] Barbierato E, Gatti A. The challenges of machine learning: a critical review. Electronics. 2024;13(2):416.

[40] Sarker IH. Machine learning: algorithms, real-world applications and research directions. SN Comput Sci. 2021;2(3):160.

[41] An Q, Rahman S, Zhou J, Kang JJ. A comprehensive review on machine learning in healthcare industry: classification, restrictions, opportunities and challenges. Sensors. 2023;23(9):4178.

[42] Krenz H, Sansone A, Fujarski M, Krallmann C, Zitzmann M, Dugas M, et al. Machine learning based prediction models in male reproductive health: development of a proof-of-concept model for Klinefelter syndrome in azoospermic patients. Andrology. 2022;10(3):534–544.

[43] Ory J, Tradewell MB, Blankstein U, Lima TF, Nackeeran S, Gonzalez DC, et al. Artificial intelligence based machine learning models predict sperm parameter upgrading after varicocele repair: a multi-institutional analysis. World J Men's Health. 2022;40(4):618–626.

[44] Gao F, Huang K, Xing Y. Artificial intelligence in omics. Genomics Insights. 2022;20(5):811–813.

[45] Shur JD, Doran SJ, Kumar S, Ap Dafydd D, Downey K, O'Connor JPB, et al. Radiomics in oncology: a practical guide. Radiographics. 2021;41(6):1717–1732.

[46] Miotto R, Wang F, Wang S, Jiang X, Dudley JT. Deep learning for healthcare: review, opportunities and challenges. Brief Bioinform. 2018;19(6):1236–1246.

[47] Yamashita R, Nishio M, Do RKG, Togashi K. Convolutional neural networks: an overview and application in radiology. Insights Imaging. 2018;9(4):611–629.

[48] Das S, Tariq A, Santos T, Kantareddy SS, Banerjee I. Recurrent neural networks (RNNs): architectures, training tricks, and introduction to influential research. In: Colliot O, editor. Machine learning for brain disorders. New York: Humana; 2023. p. 117–138.

[49] Leslie SW, Soon-Sutton TL, Khan MA. Male infertility. Male Infertility. StatPearls [Internet]. Treasure Island (FL): StatPearls Publishing; 2024 Jan.

[50] Wang C, Swerdloff RS. Limitations of semen analysis as a test of male fertility and anticipated needs from newer tests. Fertil Steril. 2014;102(6):1502–1507.

[51] Kandil H, Farkouh A, Saleh R, Boitrelle F, Agarwal A. Sperm DNA fragmentation and male infertility. In: Dahan MH, Fatemi HM, Polyzos NP, Garcia-Velasco JA, editors. Handbook of current and novel protocols for the treatment of infertility, Amsterdam, Netherlands: Academic Press; 2024. p. 29–52.

[52] Linn E, Ghanem L, Bhakta H, Greer C, Avella M. Genes regulating spermatogenesis and sperm function associated with rare disorders. Front Cell Dev Biol. 2021;9:634536.

[53] Kasimanickam V, Kasimanickam R. MicroRNAs and their associated genes regulating the acrosome reaction in sperm of high- versus low-fertility holstein bulls. Animals. 2024;14(6):833.

[54] Shamsi M, Kumar K, Dada R. Genetic and epigenetic factors: role in male infertility. Indian Journal of Urology. 2011;27(1):110.

[55] Santi D, Crépieux P, Reiter E, Spaggiari G, Brigante G, Casarini L, et al. Follicle-stimulating hormone (FSH) action on spermatogenesis: a focus on physiological and therapeutic roles. J Clin Med. 2020;9(4):1014.

[56] Kaufmann SJ, Eastaugh JL, Snowden S, Smye SW, Sharma V. The application of neural networks in predicting the outcome of in-vitro fertilization. Hum Reprod. 1997;12(7):1454–1457.

[57] Vogiatzi P, Pouliakis A, Siristatidis C. An artificial neural network for the prediction of assisted reproduction outcome. J Assist Reprod Genet. 2019;36(7):1441–1448.

[58] Yi WJ, Park KS, Paick JS. Morphological classification of sperm heads using artificial neural networks. Stud Health Technol Inform. 1998;52 Pt 2:1071–1074.

[59] Wang Y, Riordon J, Kong T, Xu Y, Nguyen B, Zhong J, et al. Prediction of DNA integrity from morphological parameters using a single-sperm DNA fragmentation index assay. Adv Sci. 2019;6(15):1900712.

[60] Kandel ME, Rubessa M, He YR, Schreiber S, Meyers S, Matter Naves L, et al. Reproductive outcomes predicted by phase imaging with computational specificity of spermatozoon ultrastructure. Proc Natl Acad Sci USA. 2020;117(31):18302–18309.

[61] Ramasamy R, Padilla WO, Osterberg EC, Srivastava A, Reifsnyder JE, Niederberger C, et al. A comparison of models for predicting sperm retrieval before microdissection testicular sperm extraction in men with nonobstructive azoospermia. J Urol. 2013;189(2):638–642.

[62] Goodson SG, White S, Stevans AM, Bhat S, Kao CY, Jaworski S, et al. CASAnova: a multiclass support vector machine model for the classification of human sperm motility patterns. Biol Reprod. 2017;97(5):698–708.

[63] Zeadna A, Khateeb N, Rokach L, Lior Y, Har-Vardi I, Harlev A, et al. Prediction of sperm extraction in non-obstructive azoospermia patients: a machine-learning perspective. Hum Reprod. 2020;35(7):1505–1514.

[64] Medina-Rodríguez R, Guzmán-Masías L, Alatrista-Salas H, Beltrán-Castañón C. Sperm cells segmentation in micrographic images through lambertian reflectance model. In: Azzopardi G, Petkov N, editors. Computer analysis of images and patterns. Cham: Springer; 2015. p. 664–674.

[65] Hidayatullah P, Zuhdi M. Automatic sperms counting using adaptive local threshold and ellipse detection. 2014 International Conference on Information Technology Systems and Innovation (ICITSI), IEEE; 2014. p. 56–61.

[66] Thiermann R, Sandler M, Ahir G, Sauls JT, Schroeder JW, Brown SD, et al. Tools and methods for high-throughput single-cell imaging with the mother machine. bioRxiv. 2024;12:1–24.

[67] Dobrovolny M, Mls K, Krejcar O, Mambou S, Selamat A. Medical image data upscaling with generative adversarial networks. In: Rojas I, Valenzuela O, Rojas F, Herrera LJ, editors. Bioinformatics and Biomedical Engineering. Cham: Springer. 2020. p. 739–749.

[68] Dobrovolny M, Benes J, Langer J, Krejcar O, Selamat A. Study on sperm-cell detection using YOLOv5 architecture with labaled dataset. Genes. 2023;14(2):451.

[69] Su TW, Choi I, Feng J, Huang K, McLeod E, Ozcan A. Sperm trajectories form chiral ribbons. Sci Rep. 2013;3(1):1664.

[70] Hidayatullah P, Mengko TLER, Munir R. A survey on multisperm tracking for sperm motility measurement. Int J Mach Learn Comput. 2017;7(5):144–151.

[71] Franco G, Scarselli F, Casciani V, De Nunzio C, Dente D, Leonardo C, et al. A novel stepwise micro-TESE approach in non obstructive azoospermia. BMC Urol. 2016;16(1):20.

[72] Ouitrakul S, Sukprasert M, Treetampinich C, Choktanasiri W, Vallibhakara S, Satirapod C. The effect of different timing after ejaculation on sperm motility and viability in semen analysis at room temperature. J Med Assoc Thai. 2018;101:26–32.

[73] Anderson RE, Hotaling JM. Inherent difficulties of meta-analysis for surgical techniques in male infertility: an argument for standardizing reporting and outcomes. Fertil Steril. 2015;104(5):1127–1128.

[74] Ramasamy R, Reifsnyder JE, Bryson C, Zaninovic N, Liotta D, Cook CA, et al. Role of tissue digestion and extensive sperm search after microdissection testicular sperm extraction. Fertil Steril. 2011;96(2):299–302.

[75] Cubuk ED, Zoph B, Mane D, Vasudevan V, Le QV. AutoAugment: learning augmentation policies from data. arXiv. 2019:113–123.

[76] Tremblay J, Prakash A, Acuna D, Brophy M, Jampani V, Ani C, et al. Training deep networks with synthetic data: bridging the reality gap by domain randomization. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops. 2018. p. 1082–1090.

[77] Beltramin D, Lamas E, Bousquet C. Ethical issues in the utilization of black boxes for artificial intelligence in medicine. Stud Health Technol Inform. 2022;295:249–252.

[78] Chen MM, Golding LP, Nicola GN. Who will pay for AI? Radiol Artif Intell. 2021;3(3):e210030.

[79] Gül M, Russo GI, Kandil H, Boitrelle F, Saleh R, Chung E, et al. Male infertility: new developments, current challenges, and future directions. World J Men's Health. 2024;42(3):502–517.

RIGHTS & PERMISSIONS

2024 The Author(s). UroPrecision published by John Wiley & Sons Australia, Ltd on behalf of Higher Education Press.