https://orcid.org/0000-0002-0757-896X
Abstract
Artificial Intelligence (AI) has witnessed significant growth in the field of medicine, leveraging machine learning, artificial neuron networks, and large language models. These technologies are effective in disease diagnosis, education, and prevention, while raising ethical concerns and potential challenges. However, their utility in sexual medicine remains relatively unexplored.
We aim to provide a comprehensive summary of the status of AI in the field of sexual medicine.
A comprehensive search was conducted using MeSH keywords, including "artificial intelligence," "sexual medicine," "sexual health," and "machine learning." Two investigators screened articles for eligibility within the PubMed and MEDLINE databases, with conflicts resolved by a third reviewer. Articles in English language that reported on AI in sexual medicine and health were included. A total of 69 full-text articles were systematically analyzed based on predefined inclusion criteria. Data extraction included information on article characteristics, study design, assessment methods, and outcomes.
The initial search yielded 905 articles relevant to AI in sexual medicine. Upon assessing the full texts of 121 articles for eligibility, 52 studies unrelated to AI in sexual health were excluded, resulting in 69 articles for systematic review. The analysis revealed AI's accuracy in preventing, diagnosing, and decision-making in sexually transmitted diseases. AI also demonstrated the ability to diagnose and offer precise treatment plans for male and female sexual dysfunction and infertility, accurately predict sex from bone and teeth imaging, and correctly predict and diagnose sexual orientation and relationship issues. AI emerged as a promising modality with significant implications for the future of sexual medicine.
Further research is essential to unlock the potential of AI in sexual medicine. AI presents advantages such as accessibility, user-friendliness, confidentiality, and a preferred source of sexual health information. However, it still lags human healthcare providers in terms of compassion and clinical expertise.
Introduction
Six decades ago, the inception of Artificial Intelligence (AI) marked a transformative era in medicine, with the evolution of machine learning (ML), artificial neuron networks (ANNs), and large language models (LLMs) leaving an indelible impact.1 The application of AI in medicine has become an indispensable tool, with extensive studies spanning diverse medical disciplines, including but not limited to the diagnosis of skin diseases,2 eye diseases,3–5 and lung cancer screening.6 Urology, as a specialized field, has been a focal point in AI research, particularly in the realm of deep learning systems for grading prostate conditions.7–10 AI has showcased its efficacy across various medical specialties, demonstrating its ability in diagnosis, education, and disease prevention, while concurrently raising ethical considerations and potential challenges.11 These encompass concerns regarding information accuracy, the perceived absence of human empathy in responses, and the need for extensive research in this evolving field.
In the domain of urology, AI has proven instrumental across various facets, including disease diagnosis, outcome predictions, treatment planning, and assessments of surgical proficiency.12 Notably, ML algorithms have significantly enhanced predictive diagnostics in conditions such as prostate cancer, renal cancer, and hydronephrosis/urinary reflux. Simultaneously, the utilization of expansive datasets plays a pivotal role in training AI algorithms for predicting outcomes in prostate cancer, urothelial issues, and urolithiasis. As a sub-discipline of urology, sexual medicine can benefit from the potential of AI and its numerous applications. With AI enhancing healthcare outcomes in various medical domains, there is potential to integrate it in sexual health and education. On brief examination, AI may offer personalized education on sexual and reproductive health (SRH), serve as a tool to screen for sexually transmitted infections (STIs), or predict sexual behaviors.13 The comprehensive utility in sexual medicine and sexual health, however, is poorly evaluated and largely unreported. This article aims to report and analyze the role and current applications of AI within the field of sexual medicine.
Material and methods
Search strategy
We conducted a detailed literature search utilizing the PubMed and Medline databases on February 9th, 2024, using the terms “artificial intelligence,” “sexual medicine,” and “sexual health.” Articles published between 1994 and 2024 were included. Ethical approval was not required for this review. We selected the scoping review methodology to map the literature on the broad topic of AI in sexual medicine, which is diverse in study design, methods, and outcomes. This approach provides a comprehensive overview of current knowledge, identifies evidence gaps, and guides future research and clinical practice.14,15
Eligibility criteria
Articles were reviewed and selected for inclusion based on relevance, without date restrictions. Inclusion criteria focused specifically on studies investigating the integration of AI in sexual medicine and health. Excluded were studies not assessing the subject, those not written in English, systematic reviews, clinical trial protocols, comments, and letters.
Screening and selection
The screening process, encompassing titles, abstracts, and full texts, was completed through Covidence (Covidence systematic review software, Veritas Health Innovation, Melbourne, Australia), a web-based software that streamlines systematic review processes by facilitating study screening, data extraction, and team collaboration. Two researchers, SL and EAC, screened, reviewed, and performed data selection and abstraction. Conflicts during the screening process were resolved by a third researcher, MAMH.
Data extraction and synthesis
Data extraction categorized articles into education, prediction, prevention, or diagnosis of sexual medicine diseases and issues. The extracted data included the study's title, design, authors, category within sexual medicine, publication year, journal published in, summary of results, and the authors' stance on AI in sexual medicine, categorized as either supportive or critical. Additionally, we summarized the main points of each article and assessed whether the authors were supportive or critical of AI integration in sexual medicine. The data were presented in a narrative format, organized into subgroups based on relevant topics. A more comprehensive overview of these studies can be found in Appendix Table 1.
Results
Overview of results
After a comprehensive review of 905 articles was conducted, a total of 69 articles were included in the final analysis (Figure 1).
Study acquisition flowchart.
The studies included were published between 1994 and 2023. However, the majority of studies were published between 2021 and 2023, with the highest number appearing in 2023. Most publications focused on testing AI models, while prospective studies were the second most common type. All studies demonstrated support for the use and benefits of AI in sexual medicine.
The largest subgroup topic of the articles reviewed (n = 21) delved into applications and evaluations aimed at preventing the spread of STIs through applications or educational initiatives. This group addressed public health challenges posed by STIs such as HIV.
The second largest subgroup (n = 20) composed of articles focusing on the use of Chatbots in a variety of subsects of sexual health, ranging from sexual dysfunction, to monkeypox.
The third subgroup (n = 9) primarily described applications or reviews focused on sexual medicine, male and/or female infertility, or education regarding sexual health habits.
The fourth subgroup (n = 9) showcased the various avenues through which AI can be employed to treat infertility, ranging from predictive models of infertility to AI-guided sperm retrieval and molecular analysis.
The final subgroup (n = 10) focused on forensics and laboratory-based studies determining the sex of bones through AI and showcasing the potential of technology in laboratory settings (Table 1).
Summary of Quantitative Data of Each Subgroup Topic.
| HIV and STIs | Chatbots | Infertility | Sex Determination and Bone Analysis | Psychology of Relationships and Sexual Behaviors |
|---|---|---|---|---|
| Model to predict internet health: Areas under the ROC curve ≥ .75 [40] | ChatGPT had a 65% accuracy in conceiving new systematic review topics33 | Effects of hirudin on max internal cavernous pressure, max ICP/mean systemic blood pressure, and area under the curve were all significantly (P < 0.01) or remarkably (P < 0.001) changed59 | 97.25% success rate in gender classification69 | Effect size for in-person infidelity for all participants was between .28 and .36 indicating a medium effect size Mean squared error: 0.82 (0.08) R2: 0.47 (0.05)74 |
| Median PrEP adherence: 91% 84% participants reported app helped with taking PrEP38 | Best use in sex education, where sensitive questions could be asked freely and openly16 | Using VGG19 network for sperm retrieval: Accuracy: 93% F1 score: 92% Average precision: 0.98 [61] | body length and body height: 75.1 % accuracy XG Boost classifier accuracy: 83 %, sensitivity: 0.81, specificity: 0.84 ROC-AUC: is 0.89 [64] | AI model predicted around 40% of variance in dyadic and solitary sexual desire75 |
| Significant reduction over time (P < 0.001) for condomless anal sex act39 | Frequency (M = 11) and duration of conversations (3:57 minutes) was high17 | 100% sensitivity, >93% specificity when differentiating Klinefelter syndrome patients from azoospermic patients57 | Accuracies for sex determination via lumbar vertebrae: Digital caliper: 82.7% Image analysis: 90.0%, Deep learning: 92.5%66 | ML model identifying individual sexual dysfunction cases in psychiatric patients: Accuracy: 0.73 Sensitivity: 0.750 Specificity: 0.721 Positive predictive value: 0.761 Negative predictive value 0.710 [73] |
| 13% (4/31) of participants perceived AI chatbots as useful for information with a simple user interface29 | Average engagement with SnehAI was 1.9 sessions, 7.6 minutes, and 56.2 messages exchanged19 | Computational models for predicting ED ➔ ROC areas for 4 hidden node network was 0.702 [58] | Training accuracy: 90% to 97%) Test set accuracy when determining sex of proximal femur: 81% to 88%93 | Frequency of PSB, as defined by screening positive on the SAST-R Core, was 8.49% in the women and 19.7% logistic regression model using these variables (pseudo R2 = 0.13) revealed that RCBS, ASRS, IAT, OCD, childhood trauma, and gender (men) remained significant78 |
| Random forest and decision tree: >80% in accuracy, precision, recall, and F1-score. Both algorithms: 80% sensitivity and specificity. AUC: 97% (random forest) and 94% (decision tree)41 | 22% perceived AI as effective, 24% as ineffective, 54% were unsure22 | Clinical decision support system in predicting ED Accuracy: 74.72%-76.65% Sensitivity: 72.33%–83.76% Specificity: 69.54%–77.10%, g-mean: 0.7468–0.7632 AUC: 0.766–0.817 [60] | Sex determination of femurs: REPTree Univariate: 60.0-87.5% accuracy REPTree Multivariate 84.0-92.5% of cases67 | Variables capturing one’s own perceptions of the relationship (e.g., conflict, affection) predicted up to 45% of the variance in relationship quality at the beginning of each study and up to 18% of the variance in relationship quality at the end of each study79 |
| Model assessing sexually risky behaviors: AUC: 78.5% to 82.8%. Boosted tree model: sensitivity of 84% (95% CI 72–92), specificity of 71% (95% CI 67–76), negative predictive value of 95% (95% CI 93–96)49 | 40% of participants found AI chatbots acceptable23 | ~70% of users favor the MenGO program76 | Gender determination using dental records: accuracy of 86.5% and 82.5% in uncropped image data group and cropped image data group respectively68 | 99.5% precision in detecting positive cases in our sample of 1 064 735 ED encounters81 |
| AUC for predicting HIV in MSM: Gradient boosting machine: 76.3% Extreme gradient boosting: 71.1%, Random forest: 72.0% Deep learning: 75.8% MLR: 69.8%45 | Dark triad traits associated with cyberbullying ➔ Random Forest method AUC: 0.983 [28] | On average, our proposed model predicting sex using ECG reached an accuracy of 94.82% ± 1.96%70 | Predicting suicidal ideation in gender and sexual minorities: 1-month follow-up mean AUC: 0.77 (95% CI, 0.74-0.79) 3-month follow-up mean AUC: 0.74 (95% CI, 0.72-0.76) 8-month follow-up mean AUC: 0.72 (95% CI, 0.69-0.74)77 | |
| ML models predicting HIV/STI testing and clinical attendance: XGBoost: AUC 62.8% (3.2%); F1 score 70.8% (1.2%)]. Elastic net regression model: AUC 82.7% (6.3%); F1 score 85.3% (1.8%)]43 | On average, there were 0.83 social and behavioral determinants of sexual health mentions per annotated note35 | Neural network to estimate sex from cranial metric traits: estimated accuracy of 84.3% and log loss of 0.348 [72] | ||
| 58.2% located in the 10 states with highest new HIV diagnoses 68.1% were in counties or states priority jurisdictions51 | DermaAId: sensitivity of 68.9% in correctly diagnosing genital diseases31 | average accuracy of 62% (±6.72) for predicting sexual orientation based on GMV and an average predictive accuracy of 92% (±9.89) using RSFC65 | ||
| Increased stimulant use associated with: STI diagnosis (1.59; 1.14–2.21)42 | Chi-square test determined that there was no significant difference in overall consultation choice between: embarrassing and stigmatizing conditions [x2 (1, 202) = 2.38, P = 0.12] sexual vs. non-sexual symptoms [x2(1, 402) = 28.17, P < 0.001]24 | |||
| Detection of occult HIV in young MSM: Ensemble approach: 96.6% accuracy and 94.6% F1 measure52 | Higher satisfaction at endline vs. baseline on reported ability to exercise sexual rights (P ≤ 0.01), confidence discussing contraception (P ≤ 0.02) and sexual feelings/needs (P ≤ 0.001) with their sexual partner(s)81 | |||
| Treatment group is the most important predictor: variable importance index of 12.9%47 | ||||
| Using social media to predict HIV and substance use: F1 scores: HIV: 82.6%, Amphetamine use: 85.9% Methamphetamine use: 85.3%53 | ||||
| Situational and contextual risk factors for HIV risk behavior among MSM not on PrEP: Random forest models: 80% specificity, 74% sensitivity44 | ||||
| Predicting HIV/STI acquisition: HIV (AUC = 0.72 Syphilis: AUC = 0.75 Gonorrhea: AUC = 0.73 Chlamydia: AUC = 0.67 [43] | ||||
| HIV prediction algorithm: LASSO: cv-AUC of 0.86 (95% CI 0.82–0.90) for identification of incident HIV infections, 0.91 (0.81–1.00) on prospective validation, and 0.77 (0.74–0.79) on external validation46 |
| HIV and STIs | Chatbots | Infertility | Sex Determination and Bone Analysis | Psychology of Relationships and Sexual Behaviors |
|---|---|---|---|---|
| Model to predict internet health: Areas under the ROC curve ≥ .75 [40] | ChatGPT had a 65% accuracy in conceiving new systematic review topics33 | Effects of hirudin on max internal cavernous pressure, max ICP/mean systemic blood pressure, and area under the curve were all significantly (P < 0.01) or remarkably (P < 0.001) changed59 | 97.25% success rate in gender classification69 | Effect size for in-person infidelity for all participants was between .28 and .36 indicating a medium effect size Mean squared error: 0.82 (0.08) R2: 0.47 (0.05)74 |
| Median PrEP adherence: 91% 84% participants reported app helped with taking PrEP38 | Best use in sex education, where sensitive questions could be asked freely and openly16 | Using VGG19 network for sperm retrieval: Accuracy: 93% F1 score: 92% Average precision: 0.98 [61] | body length and body height: 75.1 % accuracy XG Boost classifier accuracy: 83 %, sensitivity: 0.81, specificity: 0.84 ROC-AUC: is 0.89 [64] | AI model predicted around 40% of variance in dyadic and solitary sexual desire75 |
| Significant reduction over time (P < 0.001) for condomless anal sex act39 | Frequency (M = 11) and duration of conversations (3:57 minutes) was high17 | 100% sensitivity, >93% specificity when differentiating Klinefelter syndrome patients from azoospermic patients57 | Accuracies for sex determination via lumbar vertebrae: Digital caliper: 82.7% Image analysis: 90.0%, Deep learning: 92.5%66 | ML model identifying individual sexual dysfunction cases in psychiatric patients: Accuracy: 0.73 Sensitivity: 0.750 Specificity: 0.721 Positive predictive value: 0.761 Negative predictive value 0.710 [73] |
| 13% (4/31) of participants perceived AI chatbots as useful for information with a simple user interface29 | Average engagement with SnehAI was 1.9 sessions, 7.6 minutes, and 56.2 messages exchanged19 | Computational models for predicting ED ➔ ROC areas for 4 hidden node network was 0.702 [58] | Training accuracy: 90% to 97%) Test set accuracy when determining sex of proximal femur: 81% to 88%93 | Frequency of PSB, as defined by screening positive on the SAST-R Core, was 8.49% in the women and 19.7% logistic regression model using these variables (pseudo R2 = 0.13) revealed that RCBS, ASRS, IAT, OCD, childhood trauma, and gender (men) remained significant78 |
| Random forest and decision tree: >80% in accuracy, precision, recall, and F1-score. Both algorithms: 80% sensitivity and specificity. AUC: 97% (random forest) and 94% (decision tree)41 | 22% perceived AI as effective, 24% as ineffective, 54% were unsure22 | Clinical decision support system in predicting ED Accuracy: 74.72%-76.65% Sensitivity: 72.33%–83.76% Specificity: 69.54%–77.10%, g-mean: 0.7468–0.7632 AUC: 0.766–0.817 [60] | Sex determination of femurs: REPTree Univariate: 60.0-87.5% accuracy REPTree Multivariate 84.0-92.5% of cases67 | Variables capturing one’s own perceptions of the relationship (e.g., conflict, affection) predicted up to 45% of the variance in relationship quality at the beginning of each study and up to 18% of the variance in relationship quality at the end of each study79 |
| Model assessing sexually risky behaviors: AUC: 78.5% to 82.8%. Boosted tree model: sensitivity of 84% (95% CI 72–92), specificity of 71% (95% CI 67–76), negative predictive value of 95% (95% CI 93–96)49 | 40% of participants found AI chatbots acceptable23 | ~70% of users favor the MenGO program76 | Gender determination using dental records: accuracy of 86.5% and 82.5% in uncropped image data group and cropped image data group respectively68 | 99.5% precision in detecting positive cases in our sample of 1 064 735 ED encounters81 |
| AUC for predicting HIV in MSM: Gradient boosting machine: 76.3% Extreme gradient boosting: 71.1%, Random forest: 72.0% Deep learning: 75.8% MLR: 69.8%45 | Dark triad traits associated with cyberbullying ➔ Random Forest method AUC: 0.983 [28] | On average, our proposed model predicting sex using ECG reached an accuracy of 94.82% ± 1.96%70 | Predicting suicidal ideation in gender and sexual minorities: 1-month follow-up mean AUC: 0.77 (95% CI, 0.74-0.79) 3-month follow-up mean AUC: 0.74 (95% CI, 0.72-0.76) 8-month follow-up mean AUC: 0.72 (95% CI, 0.69-0.74)77 | |
| ML models predicting HIV/STI testing and clinical attendance: XGBoost: AUC 62.8% (3.2%); F1 score 70.8% (1.2%)]. Elastic net regression model: AUC 82.7% (6.3%); F1 score 85.3% (1.8%)]43 | On average, there were 0.83 social and behavioral determinants of sexual health mentions per annotated note35 | Neural network to estimate sex from cranial metric traits: estimated accuracy of 84.3% and log loss of 0.348 [72] | ||
| 58.2% located in the 10 states with highest new HIV diagnoses 68.1% were in counties or states priority jurisdictions51 | DermaAId: sensitivity of 68.9% in correctly diagnosing genital diseases31 | average accuracy of 62% (±6.72) for predicting sexual orientation based on GMV and an average predictive accuracy of 92% (±9.89) using RSFC65 | ||
| Increased stimulant use associated with: STI diagnosis (1.59; 1.14–2.21)42 | Chi-square test determined that there was no significant difference in overall consultation choice between: embarrassing and stigmatizing conditions [x2 (1, 202) = 2.38, P = 0.12] sexual vs. non-sexual symptoms [x2(1, 402) = 28.17, P < 0.001]24 | |||
| Detection of occult HIV in young MSM: Ensemble approach: 96.6% accuracy and 94.6% F1 measure52 | Higher satisfaction at endline vs. baseline on reported ability to exercise sexual rights (P ≤ 0.01), confidence discussing contraception (P ≤ 0.02) and sexual feelings/needs (P ≤ 0.001) with their sexual partner(s)81 | |||
| Treatment group is the most important predictor: variable importance index of 12.9%47 | ||||
| Using social media to predict HIV and substance use: F1 scores: HIV: 82.6%, Amphetamine use: 85.9% Methamphetamine use: 85.3%53 | ||||
| Situational and contextual risk factors for HIV risk behavior among MSM not on PrEP: Random forest models: 80% specificity, 74% sensitivity44 | ||||
| Predicting HIV/STI acquisition: HIV (AUC = 0.72 Syphilis: AUC = 0.75 Gonorrhea: AUC = 0.73 Chlamydia: AUC = 0.67 [43] | ||||
| HIV prediction algorithm: LASSO: cv-AUC of 0.86 (95% CI 0.82–0.90) for identification of incident HIV infections, 0.91 (0.81–1.00) on prospective validation, and 0.77 (0.74–0.79) on external validation46 |
Summary of Quantitative Data of Each Subgroup Topic.
| HIV and STIs | Chatbots | Infertility | Sex Determination and Bone Analysis | Psychology of Relationships and Sexual Behaviors |
|---|---|---|---|---|
| Model to predict internet health: Areas under the ROC curve ≥ .75 [40] | ChatGPT had a 65% accuracy in conceiving new systematic review topics33 | Effects of hirudin on max internal cavernous pressure, max ICP/mean systemic blood pressure, and area under the curve were all significantly (P < 0.01) or remarkably (P < 0.001) changed59 | 97.25% success rate in gender classification69 | Effect size for in-person infidelity for all participants was between .28 and .36 indicating a medium effect size Mean squared error: 0.82 (0.08) R2: 0.47 (0.05)74 |
| Median PrEP adherence: 91% 84% participants reported app helped with taking PrEP38 | Best use in sex education, where sensitive questions could be asked freely and openly16 | Using VGG19 network for sperm retrieval: Accuracy: 93% F1 score: 92% Average precision: 0.98 [61] | body length and body height: 75.1 % accuracy XG Boost classifier accuracy: 83 %, sensitivity: 0.81, specificity: 0.84 ROC-AUC: is 0.89 [64] | AI model predicted around 40% of variance in dyadic and solitary sexual desire75 |
| Significant reduction over time (P < 0.001) for condomless anal sex act39 | Frequency (M = 11) and duration of conversations (3:57 minutes) was high17 | 100% sensitivity, >93% specificity when differentiating Klinefelter syndrome patients from azoospermic patients57 | Accuracies for sex determination via lumbar vertebrae: Digital caliper: 82.7% Image analysis: 90.0%, Deep learning: 92.5%66 | ML model identifying individual sexual dysfunction cases in psychiatric patients: Accuracy: 0.73 Sensitivity: 0.750 Specificity: 0.721 Positive predictive value: 0.761 Negative predictive value 0.710 [73] |
| 13% (4/31) of participants perceived AI chatbots as useful for information with a simple user interface29 | Average engagement with SnehAI was 1.9 sessions, 7.6 minutes, and 56.2 messages exchanged19 | Computational models for predicting ED ➔ ROC areas for 4 hidden node network was 0.702 [58] | Training accuracy: 90% to 97%) Test set accuracy when determining sex of proximal femur: 81% to 88%93 | Frequency of PSB, as defined by screening positive on the SAST-R Core, was 8.49% in the women and 19.7% logistic regression model using these variables (pseudo R2 = 0.13) revealed that RCBS, ASRS, IAT, OCD, childhood trauma, and gender (men) remained significant78 |
| Random forest and decision tree: >80% in accuracy, precision, recall, and F1-score. Both algorithms: 80% sensitivity and specificity. AUC: 97% (random forest) and 94% (decision tree)41 | 22% perceived AI as effective, 24% as ineffective, 54% were unsure22 | Clinical decision support system in predicting ED Accuracy: 74.72%-76.65% Sensitivity: 72.33%–83.76% Specificity: 69.54%–77.10%, g-mean: 0.7468–0.7632 AUC: 0.766–0.817 [60] | Sex determination of femurs: REPTree Univariate: 60.0-87.5% accuracy REPTree Multivariate 84.0-92.5% of cases67 | Variables capturing one’s own perceptions of the relationship (e.g., conflict, affection) predicted up to 45% of the variance in relationship quality at the beginning of each study and up to 18% of the variance in relationship quality at the end of each study79 |
| Model assessing sexually risky behaviors: AUC: 78.5% to 82.8%. Boosted tree model: sensitivity of 84% (95% CI 72–92), specificity of 71% (95% CI 67–76), negative predictive value of 95% (95% CI 93–96)49 | 40% of participants found AI chatbots acceptable23 | ~70% of users favor the MenGO program76 | Gender determination using dental records: accuracy of 86.5% and 82.5% in uncropped image data group and cropped image data group respectively68 | 99.5% precision in detecting positive cases in our sample of 1 064 735 ED encounters81 |
| AUC for predicting HIV in MSM: Gradient boosting machine: 76.3% Extreme gradient boosting: 71.1%, Random forest: 72.0% Deep learning: 75.8% MLR: 69.8%45 | Dark triad traits associated with cyberbullying ➔ Random Forest method AUC: 0.983 [28] | On average, our proposed model predicting sex using ECG reached an accuracy of 94.82% ± 1.96%70 | Predicting suicidal ideation in gender and sexual minorities: 1-month follow-up mean AUC: 0.77 (95% CI, 0.74-0.79) 3-month follow-up mean AUC: 0.74 (95% CI, 0.72-0.76) 8-month follow-up mean AUC: 0.72 (95% CI, 0.69-0.74)77 | |
| ML models predicting HIV/STI testing and clinical attendance: XGBoost: AUC 62.8% (3.2%); F1 score 70.8% (1.2%)]. Elastic net regression model: AUC 82.7% (6.3%); F1 score 85.3% (1.8%)]43 | On average, there were 0.83 social and behavioral determinants of sexual health mentions per annotated note35 | Neural network to estimate sex from cranial metric traits: estimated accuracy of 84.3% and log loss of 0.348 [72] | ||
| 58.2% located in the 10 states with highest new HIV diagnoses 68.1% were in counties or states priority jurisdictions51 | DermaAId: sensitivity of 68.9% in correctly diagnosing genital diseases31 | average accuracy of 62% (±6.72) for predicting sexual orientation based on GMV and an average predictive accuracy of 92% (±9.89) using RSFC65 | ||
| Increased stimulant use associated with: STI diagnosis (1.59; 1.14–2.21)42 | Chi-square test determined that there was no significant difference in overall consultation choice between: embarrassing and stigmatizing conditions [x2 (1, 202) = 2.38, P = 0.12] sexual vs. non-sexual symptoms [x2(1, 402) = 28.17, P < 0.001]24 | |||
| Detection of occult HIV in young MSM: Ensemble approach: 96.6% accuracy and 94.6% F1 measure52 | Higher satisfaction at endline vs. baseline on reported ability to exercise sexual rights (P ≤ 0.01), confidence discussing contraception (P ≤ 0.02) and sexual feelings/needs (P ≤ 0.001) with their sexual partner(s)81 | |||
| Treatment group is the most important predictor: variable importance index of 12.9%47 | ||||
| Using social media to predict HIV and substance use: F1 scores: HIV: 82.6%, Amphetamine use: 85.9% Methamphetamine use: 85.3%53 | ||||
| Situational and contextual risk factors for HIV risk behavior among MSM not on PrEP: Random forest models: 80% specificity, 74% sensitivity44 | ||||
| Predicting HIV/STI acquisition: HIV (AUC = 0.72 Syphilis: AUC = 0.75 Gonorrhea: AUC = 0.73 Chlamydia: AUC = 0.67 [43] | ||||
| HIV prediction algorithm: LASSO: cv-AUC of 0.86 (95% CI 0.82–0.90) for identification of incident HIV infections, 0.91 (0.81–1.00) on prospective validation, and 0.77 (0.74–0.79) on external validation46 |
| HIV and STIs | Chatbots | Infertility | Sex Determination and Bone Analysis | Psychology of Relationships and Sexual Behaviors |
|---|---|---|---|---|
| Model to predict internet health: Areas under the ROC curve ≥ .75 [40] | ChatGPT had a 65% accuracy in conceiving new systematic review topics33 | Effects of hirudin on max internal cavernous pressure, max ICP/mean systemic blood pressure, and area under the curve were all significantly (P < 0.01) or remarkably (P < 0.001) changed59 | 97.25% success rate in gender classification69 | Effect size for in-person infidelity for all participants was between .28 and .36 indicating a medium effect size Mean squared error: 0.82 (0.08) R2: 0.47 (0.05)74 |
| Median PrEP adherence: 91% 84% participants reported app helped with taking PrEP38 | Best use in sex education, where sensitive questions could be asked freely and openly16 | Using VGG19 network for sperm retrieval: Accuracy: 93% F1 score: 92% Average precision: 0.98 [61] | body length and body height: 75.1 % accuracy XG Boost classifier accuracy: 83 %, sensitivity: 0.81, specificity: 0.84 ROC-AUC: is 0.89 [64] | AI model predicted around 40% of variance in dyadic and solitary sexual desire75 |
| Significant reduction over time (P < 0.001) for condomless anal sex act39 | Frequency (M = 11) and duration of conversations (3:57 minutes) was high17 | 100% sensitivity, >93% specificity when differentiating Klinefelter syndrome patients from azoospermic patients57 | Accuracies for sex determination via lumbar vertebrae: Digital caliper: 82.7% Image analysis: 90.0%, Deep learning: 92.5%66 | ML model identifying individual sexual dysfunction cases in psychiatric patients: Accuracy: 0.73 Sensitivity: 0.750 Specificity: 0.721 Positive predictive value: 0.761 Negative predictive value 0.710 [73] |
| 13% (4/31) of participants perceived AI chatbots as useful for information with a simple user interface29 | Average engagement with SnehAI was 1.9 sessions, 7.6 minutes, and 56.2 messages exchanged19 | Computational models for predicting ED ➔ ROC areas for 4 hidden node network was 0.702 [58] | Training accuracy: 90% to 97%) Test set accuracy when determining sex of proximal femur: 81% to 88%93 | Frequency of PSB, as defined by screening positive on the SAST-R Core, was 8.49% in the women and 19.7% logistic regression model using these variables (pseudo R2 = 0.13) revealed that RCBS, ASRS, IAT, OCD, childhood trauma, and gender (men) remained significant78 |
| Random forest and decision tree: >80% in accuracy, precision, recall, and F1-score. Both algorithms: 80% sensitivity and specificity. AUC: 97% (random forest) and 94% (decision tree)41 | 22% perceived AI as effective, 24% as ineffective, 54% were unsure22 | Clinical decision support system in predicting ED Accuracy: 74.72%-76.65% Sensitivity: 72.33%–83.76% Specificity: 69.54%–77.10%, g-mean: 0.7468–0.7632 AUC: 0.766–0.817 [60] | Sex determination of femurs: REPTree Univariate: 60.0-87.5% accuracy REPTree Multivariate 84.0-92.5% of cases67 | Variables capturing one’s own perceptions of the relationship (e.g., conflict, affection) predicted up to 45% of the variance in relationship quality at the beginning of each study and up to 18% of the variance in relationship quality at the end of each study79 |
| Model assessing sexually risky behaviors: AUC: 78.5% to 82.8%. Boosted tree model: sensitivity of 84% (95% CI 72–92), specificity of 71% (95% CI 67–76), negative predictive value of 95% (95% CI 93–96)49 | 40% of participants found AI chatbots acceptable23 | ~70% of users favor the MenGO program76 | Gender determination using dental records: accuracy of 86.5% and 82.5% in uncropped image data group and cropped image data group respectively68 | 99.5% precision in detecting positive cases in our sample of 1 064 735 ED encounters81 |
| AUC for predicting HIV in MSM: Gradient boosting machine: 76.3% Extreme gradient boosting: 71.1%, Random forest: 72.0% Deep learning: 75.8% MLR: 69.8%45 | Dark triad traits associated with cyberbullying ➔ Random Forest method AUC: 0.983 [28] | On average, our proposed model predicting sex using ECG reached an accuracy of 94.82% ± 1.96%70 | Predicting suicidal ideation in gender and sexual minorities: 1-month follow-up mean AUC: 0.77 (95% CI, 0.74-0.79) 3-month follow-up mean AUC: 0.74 (95% CI, 0.72-0.76) 8-month follow-up mean AUC: 0.72 (95% CI, 0.69-0.74)77 | |
| ML models predicting HIV/STI testing and clinical attendance: XGBoost: AUC 62.8% (3.2%); F1 score 70.8% (1.2%)]. Elastic net regression model: AUC 82.7% (6.3%); F1 score 85.3% (1.8%)]43 | On average, there were 0.83 social and behavioral determinants of sexual health mentions per annotated note35 | Neural network to estimate sex from cranial metric traits: estimated accuracy of 84.3% and log loss of 0.348 [72] | ||
| 58.2% located in the 10 states with highest new HIV diagnoses 68.1% were in counties or states priority jurisdictions51 | DermaAId: sensitivity of 68.9% in correctly diagnosing genital diseases31 | average accuracy of 62% (±6.72) for predicting sexual orientation based on GMV and an average predictive accuracy of 92% (±9.89) using RSFC65 | ||
| Increased stimulant use associated with: STI diagnosis (1.59; 1.14–2.21)42 | Chi-square test determined that there was no significant difference in overall consultation choice between: embarrassing and stigmatizing conditions [x2 (1, 202) = 2.38, P = 0.12] sexual vs. non-sexual symptoms [x2(1, 402) = 28.17, P < 0.001]24 | |||
| Detection of occult HIV in young MSM: Ensemble approach: 96.6% accuracy and 94.6% F1 measure52 | Higher satisfaction at endline vs. baseline on reported ability to exercise sexual rights (P ≤ 0.01), confidence discussing contraception (P ≤ 0.02) and sexual feelings/needs (P ≤ 0.001) with their sexual partner(s)81 | |||
| Treatment group is the most important predictor: variable importance index of 12.9%47 | ||||
| Using social media to predict HIV and substance use: F1 scores: HIV: 82.6%, Amphetamine use: 85.9% Methamphetamine use: 85.3%53 | ||||
| Situational and contextual risk factors for HIV risk behavior among MSM not on PrEP: Random forest models: 80% specificity, 74% sensitivity44 | ||||
| Predicting HIV/STI acquisition: HIV (AUC = 0.72 Syphilis: AUC = 0.75 Gonorrhea: AUC = 0.73 Chlamydia: AUC = 0.67 [43] | ||||
| HIV prediction algorithm: LASSO: cv-AUC of 0.86 (95% CI 0.82–0.90) for identification of incident HIV infections, 0.91 (0.81–1.00) on prospective validation, and 0.77 (0.74–0.79) on external validation46 |
Chatbots
Use of AI chatbots has surged over the past decade across countless industries, including major search engines and telecommunication. Sexual health is no stranger to chatbots - numerous studies have studied the development of chatbots to assist with health services, as well as their reception by users.
Chatbots and sexual education
Studies on the use of chatbots in SRH reveal varied findings.16–28 Chatbots have shown promise in providing accessible, engaging, and anonymous information on sexual health, offering support for STI testing, education, and counseling.16,17 They have also been effective in improving attitudes and confidence related to sexual activity, as seen in tools like Sexpert and SnehAI.18,19 However, chatbots often lack empathy compared to human interactions, which is a noted limitation.
Preferences for chatbot use are influenced by stigma, convenience, and the anonymity they offer, particularly for sensitive topics.20,21 Despite their utility, hesitancy exists among healthcare professionals regarding patient safety and reliability.22 Younger individuals and those with STI-related concerns are more open to digital health solutions, although most still prefer face-to-face consultations for accuracy and reassurance.23,24 Advanced AI, such as ChatGPT, has demonstrated potential in raising awareness about sexual consent and preventing sexual violence but is not yet suitable as a sole educational tool.26,27 AI has also been successfully used to analyze relationships between psychological traits and sexual behaviors, further demonstrating its role in SRH research.28
Chatbots and STI
Chatbots have also been utilized to assist with STI testing, diagnosis, and outreach; especially for marginalized and underserved populations.29–36 For example, a chatbot developed in Malaysia catered specifically to men who have sex with men (MSM) to overcome barriers in HIV prevention and screening. This chatbot was valued for its emotional support, low stigma, and ease of use, with 19% of participants specifically highlighting its personal touch.29 Similarly, TelePrEP, another AI-driven tool, streamlined access to pre-exposure prophylaxis (PrEP) by facilitating virtual appointments, lab testing, and medication delivery, garnering positive feedback for its user-friendly approach.30
In global health crises like monkeypox, AI chatbots have been employed to analyze social media narratives and provide information. However, limitations such as misinformation and generic responses highlighted the need for ongoing improvement and careful regulation in clinical applications.32,34 Tools like Amanda Selfie, a chatbot with a Black transgender woman avatar, further emphasize the importance of relatability in fostering engagement and education, though users pointed out limitations in response quality and flexibility.34
Moreover, AI chatbots have been incorporated into educational initiatives, such as SRH modules for adolescents in Indonesia, which effectively improved knowledge and engagement through iterative user feedback. Despite their potential, challenges like inadequate coding of social and behavioral determinants in electronic health records and the need for enhanced personalization persist.35,36 These findings underscore the importance of balancing chatbot efficiency with human oversight to ensure comprehensive, accurate, and empathetic care in sexual health.36
HIV and STIs
HIV remains a significant global health challenge, disproportionately affecting racial and sexual minorities and geographic regions such as the southern United States.37 Despite advances in genomic sequencing and antiretroviral therapy, addressing these disparities requires targeted prevention and treatment strategies. AI has demonstrated substantial potential in supporting HIV-related interventions through innovative approaches in prevention, testing, and treatment.38–41
Studies highlight the use of AI-driven tools to improve adherence to preventive measures like PrEP, increase HIV testing among stigmatized populations, and support personalized treatment strategies. For example, mobile apps tailored for gay men boosted PrEP adherence,38 interventions effectively targeted homeless youth to impart HIV-related knowledge,39 and culturally sensitive chatbots effectively encouraged HIV testing among MSM in Malaysia.40 AI predictive models and ML algorithms have also been utilized to optimize online prevention efforts and analyze transmission clusters through genetic data.41 These findings collectively illustrate AI's capacity to address HIV prevention and care challenges, highlighting the importance of culturally tailored, accessible, and data-driven interventions.
HIV prevention
HIV transmission patterns, through the assistance of AI tools, have been increasingly well-defined.38,39,42 For instance, in a study ranging from 2014-2020, participants were regularly tested in 6-month intervals for STIs and surveyed for demographics, substance use, risky sexual behaviors, and partnership status. This data was used to generate a multivariate linear regression, which showed that increased stimulant use was positively associated with increased rates of STI diagnosis and transactional sex.42
aDOT, a mobile phone app which, by monitoring audio and video, acts as a sexual diary, tracking and monitoring not only sexual behaviors and encounters, but also ensuring proper PrEP adherence. In a study by Liu et al, 84% of MSM reported that aDOT was helpful with PrEP adherence.38 Additionally, using AI to select peer-change agents (PCA) at drop-in centers led to a 33% reduction in condomless anal sex (CAS) over the span of 1 month, while PCA selected by hand showed no reductions in CAS until 3 months.39
HIV testing/diagnosis: Modeling
AI and ML models have demonstrated significant promise in advancing HIV risk prediction, prevention, and testing.43–50 Studies such as Xu et al's have utilized large datasets to develop predictive models for HIV and other STIs, yielding AUC scores of 0.67 to 0.75 for conditions such as HIV, syphilis, gonorrhea, and chlamydia.43 These tools offer valuable insights into individual risk levels, facilitating targeted prevention efforts. AI algorithms applied to behavioral data like patient diaries and surveys, revealed the influence of psychological states on sexual habits, offering innovative ways to engage users in risk reduction interventions.44
AI has also been instrumental in improving the efficiency and accessibility of HIV testing.45 EHR-derived algorithms identified potential candidates for PrEP with high specificity, screening thousands to pinpoint individuals at high risk for HIV.46,47 Majam et al and Xu et al showed the efficacy of ML-driven reminder systems, such as email alerts, in promoting timely clinic attendance and HIV/STI testing, with tailored frequency and modality enhancing participation rates. These applications demonstrate the potential for AI to optimize resource allocation and patient engagement.48,50
In resource-limited settings like South Africa, ML-generated risk assessment panels provide cost-effective alternatives to large-scale HIV screening, enabling targeted counseling and testing among high-risk groups. Additionally, demographic and behavioral data have been used to create predictive models that inform interventions.49
Despite these advancements, the efficacy of AI in HIV care is not without limitations. Challenges include the inability of some models to address abstract disparities and the need for ongoing adaptation to diverse populations and behaviors.43 Nonetheless, the integration of AI and ML into HIV prevention, testing, and treatment represents a transformative step forward, with the potential to reduce the global burden of HIV through precision-targeted strategies.
HIV and social media
AI and ML applications in social media and public health have yielded groundbreaking results in HIV prevention and sexual health interventions.51–54 Social media platforms like X (formerly Twitter) have been used to analyze over 3 million tweets from users who are homosexual, bisexual, or identify as MSM. Through natural language processing, researchers have developed pipelines with a precision rate of 0.85 to identify at-risk individuals for HIV.51 Convolutional neural networks models have also been used to analyze social network data, achieving 96.6% accuracy in modeling infectious disease transmission among young MSM, based on their social, sexual, and health-related interactions.52 Furthermore, Gay Social Networking Analysis Program employed ML techniques to link social media data with offline behaviors, successfully predicting HIV status and substance use outcomes.53
Finally, we reviewed and summarized peer-reviewed publications in which AI techniques were applied to four HIV prevention-intervention objectives designed to (1) enhance the prevention of HIV infections (Prevention), (2) test and diagnose HIV infection as early as possible (Testing), (3) use treatment as prevention (Treatment), and (4) respond to emerging clusters (Response). The authors presented a structured approach for developing HIV prevention AI algorithms. It begins with clearly defining the study's objective related to the HIV intervention, followed by dataset introduction, including data collection details, variable description, and descriptive analysis. The algorithm implementation phase encompasses defining the problem, preprocessing the data, and selecting appropriate methods, including software and external packages. Evaluation strategies involve analyzing results using metrics and comparison techniques, while reproducibility focuses on ensuring code and data availability for future use or verification.54
AI and ML techniques in HIV prevention and sexual health promotion represent a challenge in public health research and intervention. Through diverse studies examining the development and implementation of AI-driven interventions such as mobile applications, chatbots, and predictive models, significant strides have been made in enhancing adherence to PrEP, disseminating HIV-related knowledge, and encouraging HIV testing among at-risk populations. These findings collectively underscore the transformative potential of AI and ML in revolutionizing strategies for combating HIV transmission and promoting sexual health awareness and behavior change.
Infertility
AI has significantly influenced the field of infertility, offering transformative potential in both assisted reproductive technologies (ART) and andrology.55–62 Recent advances in AI have automated processes like gamete quality assessment, sperm selection for intracytoplasmic sperm injection, and embryo selection, optimizing ART protocols and clinical workflows through predictive maintenance and continuous quality control.55 AI’s application extends to male infertility diagnosis, where ML algorithms help clinicians analyze vast datasets – including clinical, environmental, lifestyle, and molecular “omics” data – to predict diagnoses and tailor treatments more effectively.56 In male infertility, AI and ML algorithms have proven valuable in analyzing complex datasets that include clinical, environmental, lifestyle, and molecular data to predict diagnoses and tailor personalized treatments. For example, AI models have demonstrated high sensitivity and specificity in diagnosing genetic conditions like Klinefelter Syndrome in azoospermic patients, showcasing their ability to outperform non-expert physicians in identifying underdiagnosed conditions.57 Additionally, AI has been used to analyze sexual health issues such as erectile dysfunction (ED), with studies indicating strong correlations between ED and factors like patient age and depression, enabling better risk stratification and treatment management.58
Risk stratifying patients through early detection and diagnosis can improve prognosis and treatment in ED, genetic sexual dysfunction, infertility, and STIs.59,60 Early studies by Chen et al. focused on developing a clinical decision support system in predicting ED.60
In forensics, AI-based deep CNNs have been employed to automate sperm detection on microscope slides, achieving over 90% accuracy in sperm identification, although expert verification remains necessary to address the risk of false negatives in critical cases.61
Despite these advances, challenges persist, particularly in the implementation of AI in SRH education and decision-making tools. For example, Sütcüoğlu et al.'s evaluation of ChatGPT in addressing premature ovarian insufficiency revealed that while AI can provide mostly accurate medical advice, it also generated 20% inappropriate or inaccurate information, highlighting the limitations of AI as a standalone resource.62 Collectively, these findings underscore the transformative potential of AI in infertility management while also pointing to the need for further refinement to fully realize its benefits in clinical practice.
Sex determination and bone analysis
The integration of AL and ML technologies has reshaped both forensic sciences and medical diagnostics, offering potent tools for sex determination and gender classification.63–72 AI has been used to analyze anatomical features such as the hyoid bone for gender detection, with promising results.64 ML models have also been employed to predict sexual orientation from brain imaging, demonstrating notable accuracy.65 In forensic contexts, AI applications like lumbar imaging analysis and CT scans of the proximal femur provide effective solutions, particularly when traditional methods are unavailable.66 Other techniques, such as femur length measurements67 and gender determination from intraoral photographs,68,69,72 highlight AI's growing role in forensic investigations. Additionally, the use of panoramic dental radiographs and RGB wavelet transformation of ECG has further advanced forensic and personal identification research.70
In the realm of medical diagnostics, ML has been utilized to analyze brain patterns through MRI data, revealing links to specific risks, such as child sexual abuse.71 These diverse studies collectively highlight the transformative potential of AI and ML in advancing our understanding and capabilities in sex determination and gender classification across both forensic sciences and medical diagnostics. This convergence of technology and healthcare promises a new era of precision and efficiency in these critical domains.
Psychology of relationships and sexual Behaviors
ML studies offer valuable insights into sexual behavior, relationship dynamics, and healthcare interventions, revealing predictive risk factors for sexual dysfunction, sexual behavior, while also demonstrating the efficacy of AI in SRH education.73–81 For instance, ML has been used to detect sexual dysfunction in psychiatric patients, revealing high prevalence rates and identifying predictive risk factors.73 Other studies have explored interpersonal dynamics, such as predicting online and in-person infidelity, where factors like relationship satisfaction and love were found to be significant predictors.74,75 ML algorithms have also been applied to predict sexual desire and create tailored healthcare interventions, such as MenGO, an AI-guided digital health program designed to address ED, infertility, and other andrological issues.76
In the context of mental health, studies have utilized ML to predict suicidal ideation in gender and sexual minority populations, emphasizing the importance of contextual risk factors, such as stress levels and mood.77 Similarly, research into problematic sexual behavior (PSB) has identified psychological and environmental factors like OCD, childhood trauma, and reward deficiency syndrome as risk factors.78 Additionally, ML has been used to analyze relationship quality, demonstrating how individual and partner experiences shape relationship dynamics.79
AI technologies also play a role in trauma-informed care. For example, the Ube app enables individuals to anonymously share data about their sexual experiences on online dating platforms, with safeguards to prevent re-traumatization.80 Natural language processing has been used in emergency departments to identify intimate partner violence, achieving high precision rates in detecting such cases.81 These studies collectively highlight the diverse applications and insights gained from ML in understanding sexual behavior, relationship dynamics, and healthcare interventions.
Discussion
Key findings
One of the most prevalent debates in modern medicine is whether AI should be integrated into clinical practice or not.82 The integration of AI in medicine, especially in the fields of urology and sexual medicine, indicates a new era of patient care characterized by enhanced diagnostic accuracy,56 improved treatment adherence,38 and more effective patient follow-up.40 AI's prevalence in healthcare is driven by its capacity to process and analyze vast amounts of data quickly and accurately, offering benefits that extend from everyday clinical practice to specialized medical fields.
In the realm of sexual medicine, AI-driven digital platforms, such as smartphone applications, provide significant advantages for individuals managing STIs, including HIV.13,83 These platforms enhance medication adherence, particularly for those on PrEP by offering timely reminders and personalized support systems. This improves patient outcomes by ensuring consistent medication use, which is crucial for the effectiveness of treatments like PrEP.30,38 Additionally, these digital tools provide easier and more reliable access to medical advice compared to traditional internet searches, which are often fraught with misinformation. By leveraging sophisticated algorithms, AI applications can deliver personalized, accurate, and up-to-date health information, thus empowering patients to make informed decisions about their sexual health.16,17,40
AI also plays a critical role in the early detection and prevention of diseases. By analyzing diverse data sources, including medical tests, lifestyle factors, and patient-reported outcomes, AI can identify patterns and signals indicative of potential health issues earlier than traditional methods.56,58,59,73 This capability not only enhances the precision of diagnoses but also enables timely interventions that can prevent the progression of diseases, ultimately reducing the burden of chronic conditions. Moreover, the diagnostic accuracy of AI systems often surpasses that of conventional diagnostic methods due to their ability to handle and interpret complex datasets.56 This is particularly beneficial in sexual medicine, where precise diagnoses are essential for effective treatment planning and disease management.
One of the notable advantages of AI in sexual medicine is its potential to reduce the stigma associated with seeking sexual health advice. The anonymity and accessibility of AI-driven platforms make patients more comfortable discussing sensitive issues, leading to increased engagement and better health outcomes.24,29,84 Furthermore, AI facilitates remote monitoring, allowing for continuous care and better follow-up, which is especially important for managing chronic conditions such as HIV. This remote capability ensures that patients receive consistent monitoring and support, enhancing the overall quality of care.38,76 Moreover, the healthcare ecosystem is increasingly recognizing the significance of AI-powered tools, which can enhance various processes and significantly reduce costs. By 2026, AI applications are estimated to cut annual US healthcare costs by $150 billion.85
Challenges of AI in sexual medicine
However, the increasing use of AI in healthcare also brings significant challenges and ethical considerations. One major concern is the accuracy of the information provided by AI applications and the potential spread of misinformation.86 While AI can enhance healthcare delivery, it is crucial to ensure that the data and algorithms used are accurate and reliable.87 AI systems must be designed to flag and report inaccurate content, contributing to improved information quality and patient safety.
The potential for AI to replace physicians and healthcare providers is another concern. Despite the numerous benefits of AI, its lack of human empathy remains a significant limitation. While AI can provide data-driven insights and clinical decision recommendations, it cannot replace the human compassion, judgment, and nuanced understanding that are essential in patient care.16 The absence of these qualities in AI-driven interactions may impact patient satisfaction and the overall therapeutic experience, highlighting the importance of maintaining a human touch in healthcare. Therefore, the role of AI should be seen as complementary to that of healthcare professionals, enhancing their ability to provide high-quality care rather than replacing them. Also, ensuring that urologists retain their decision-making authority and clinical autonomy is essential for the ethical integration of AI technologies into practice.11
Privacy and confidentiality are critical issues in the use of AI in sexual medicine. The sensitive nature of sexual health information necessitates stringent measures to protect patient data from breaches and unauthorized access.88 Ensuring the secure handling of this information is paramount to maintaining patient trust and complying with legal and ethical standards. The European Union is tackling AI and data protection through existing regulations like the GDPR and new laws such as the AI Act. The GDPR already controls personal data processing and safeguards individuals from fully automated decisions, especially in healthcare. The AI Act introduces a risk-based framework to enhance ethical AI use and is expected to complement the GDPR, with many of its goals achievable before full implementation.89
Additionally, AI algorithms can perpetuate existing biases in healthcare if not carefully designed and monitored. Studies often focus on specific populations or settings, limiting the generalizability of findings. Also, language and geographic biases in existing studies may also limit the global representation of findings.90 Addressing these biases is essential to ensure equitable and unbiased care for all patients, regardless of their background or circumstances. Moreover, the rapid advancement of AI technology can render some studies outdated or less relevant, necessitating continuous research to keep pace with technological developments and assess their long-term implications. Additionally, broadening research to include diverse demographics and clinical contexts can enhance the applicability of AI solutions.91
The regulatory and legal landscape surrounding AI in healthcare is still evolving. Robust frameworks are needed to ensure the safe and effective use of AI technologies, addressing issues such as liability, accountability, and the ethical use of patient data.92 As AI continues to develop, it is crucial to establish clear guidelines and standards to govern its implementation and use in clinical practice.
Limitations
While this scoping review offers valuable insights into the applications of AI in sexual medicine, several limitations must be acknowledged. The heterogeneity of the included studies, which cover a wide range of AI techniques, datasets, and clinical applications, presents challenges in making direct comparisons and drawing standardized conclusions. Our decision to focus exclusively on literature published in English may introduce a potential bias, as it could result in the omission of valuable insights from studies published in other languages. Additionally, many studies are exploratory in nature with limited longitudinal follow-up, restricting our understanding of the long-term efficacy and safety of AI in clinical practice. The rapid pace of AI advancements and publication presents a significant challenge, as new research emerges almost daily. To effectively evaluate its integration into urology and healthcare as a whole, assessments would need to occur at more frequent intervals to stay current with ongoing developments. Addressing these limitations will require more rigorous and inclusive research, along with standardized protocols and a greater emphasis on ethical considerations.
Conclusion
In conclusion, while AI holds significant promise for transforming sexual medicine, it is essential to address its limitations and ethical considerations to ensure its successful integration into clinical practice. Increasing AI literacy among healthcare providers and the general population is crucial to harnessing the full potential of these technologies. Developing robust ethical and regulatory frameworks will support the responsible use of AI, ultimately enhancing patient care and advancing the field of sexual medicine.
Acknowledgments
None.
Author contributions
EAC & SL conceptualized the study, retrieved, and screened articles, and wrote/edited the manuscript. BI & TH extracted results and wrote/edited the manuscript. MAMH & HC conceptualized the study, wrote/edited the manuscript. FAY contributed to project conception/design and manuscript writing/editing. All authors read and approved the final manuscript.
Funding
None.
Conflicts of interest
YAF: Coloplast: Advisory board, speaker; Endo: Advisory board; Haleon: Advisory board; Halozyme: Advisory board, speaker; Masimo: Intellectual property; Softwave: Advisory board; Sprout: Consultant; Vertica: Research investigator; Xialla: Advisory board.
Data availability
Not applicable.
Ethics approval and consent to participate
Not Applicable.
Consent for publication
Not Applicable.
