Advancements in artificial intelligence for robotic-assisted radical prostatectomy in men suffering from prostate cancer: results from a scoping review

Introduction

Background

Prostate cancer (PCa) is the second cancer for incidence in men (1) with an increasing incidence rate of 2–3% annually (2). Robotic-assisted radical prostatectomy (RARP) is currently a first-line treatment option for men with localized disease, at least 10 years of life expectancy, and candidate for curative treatment (3). RARP has gained popularity in the last decade over traditional open radical prostatectomy thanks to the seamless integration of high-resolution three-dimensional magnified vision, a wide range of movements, tremor dampening, instrument miniaturization, and other features that collectively enhance the dexterity and precision of the operating surgeon’s hands, facilitating safe and efficient surgery (4). This converts into fewer immediate postoperative complications such as blood loss, transfusion rates, and shorter hospital stays (5). Yet, studies have shown slightly improved early functional outcomes in terms of urinary continence and erectile function with RARP compared to open surgery (5). This may be attributed to the enhanced visualization and precision afforded by robotic-assisted techniques which help in preserving anatomical structure such as in the Retzius sparing technique (6).

Rationale

The advancement of artificial intelligence (AI) has been progressing swiftly in recent years, encompassing enhancements in software algorithms, hardware execution, and widespread applications across numerous domains. The application of AI in medicine is still in its early stage, it holds significant promise, with many recent applications demonstrating its potential (7), it holds the potential to transform both the teaching and practice of surgery, offering the prospect of a future tailored to optimize patient care quality to its highest potential through the use of machine learning, computer vision, artificial neural networks, and natural language processing (8). The integration of AI technologies into RARP procedures might open new avenues for enhancing surgical outcomes, patient safety, and operative efficiency.

Objective

In this paper, we aim to present a comprehensive review of the current state-of-the-art applications of AI in RARP, encompassing preoperative planning, intraoperative assistance, and postoperative outcomes. We present this article in accordance with the PRISMA-ScR reporting checklist (available at https://cco.amegroups.com/article/view/10.21037/cco-24-52/rc) (9).

Methods

Aim of the study and literature search

In this study, we performed a scoping review to evaluate the role of AI on RARP for PCa. Literature search was performed on 2nd February 2024 using EMBASE, PubMed, and Scopus. The following terms and Boolean operators were used: (artificial intelligence OR machine learning OR deep learning) AND (surgical planning OR preoperative planning OR surgical outcomes) AND (robotic radical prostatectomy OR robot-assisted radical prostatectomy OR robotic-assisted radical prostatectomy). No date limit was imposed. This review was registered on https://osf.io/registries/ (number osf.io/p79mv; https://osf.io/txg3f).

Selection criteria

Only English papers were accepted. Pediatric studies were excluded. Reviews, letters to the editor, case reports, and meeting abstracts were also excluded.

Study screening and selection

Our main objective was to evaluate the application and role of AI for RARP. The PICOS (Patient Intervention Comparison Outcome Study type) model was used to frame and answer the clinical question; P: adult men with PCa undergoing RARP; I: use of AI; C: none; O: preoperative planning improvement and postoperative outcomes; S: prospective and retrospective studies. Two independent authors screened all retrieved studies through Covidence Systematic Review Management^® (Veritas Health Innovation, Melbourne, Australia). A third author solved discrepancies. The full text of the screened papers was selected if found pertinent to the purpose of this review.

Results

Literature screening

The literature search found 1,211 papers. Three papers were added from other sources. Ten duplicates were automatically excluded, and 1,204 papers remained for screening against title and abstract. Among them, 1,183 papers were further excluded because were unrelated to this review purpose. The remaining 21 full-text papers were screened for pertinence. Four papers were excluded. Ultimately, 17 papers were accepted and included (10-26). Figure 1 shows the flow diagram of the literature search.

Figure 1 PRISMA flow diagram of the flow of information through the different phases of screening.

Study characteristics

Table 1 shows study characteristics. There were four studies dealing with the prediction of positive surgical margins (PSM)/extraprostatic extension (11,17,22,26) and one with biochemical recurrence (BCR) (19), five studies with patient’s outcomes (i.e., length of stay, recovery of erectile function and urinary continence) (13,16,20,21,25), while three studies reported outcomes on application on intraoperative superimposition of magnetic resonance images to identify and locate lesions for nerve-sparing surgery (10,14,24). The remaining ones reported the use of AI in the identification and labeling of surgical steps (15), identification of the intraoperative source of bleeding (18), and quality of surgery (12,23).

Prediction of PSM, extraprostatic extension, and BCR

Hao et al. examined a population of 903 patients undergoing RARP, with 151 cases of PSM (26). They reported a statistically significant difference in age, percent of positive needles, International Society of Urological Pathology (ISUP) group, pathological stage, Prostate Imaging Reporting & Data System (PI-RADS) lesion, tumor location, maximal tumor diameter, clinical stage at magnetic resonance imaging (MRI), prostate specific antigen (PSA), and PSA-density, between the positive and negative surgical margin groups. They used a Lasso method for variable screening and logistic regression analysis to establish the final model with good performance (c-statistic 0.727) to predict the presence of PSM post-RARP. Their model is available at https://doctor-h.shinyapps.io/dynnomapp/.

Lee et al. analyzed a population of 236 patients undergoing RARP to develop a prediction model for PSM based on automated performance metrics (APMs) (validated objective measures of surgeon performance from 11 standardized RARP steps) and clinical parameters (11). The PSM rate was 23.3%, with a statistically significant difference of pT ≥3 tumors between patients with PSM and those with negative margins (83.6% vs. 48.1%, P<0.001). The full model, including patient clinical factors and APMs, showed an area under the curve (AUC) of 0.74 in predicting PSM. When assessing patient clinical factors alone or APMs alone, the model achieved AUC of 0.72 and 0.64, respectively. The most important predictors of PSM were extracapsular extension and pT stage followed by APMs. All the APMS were associated with either the bladder neck dissection (4/6; 66.7%) or lymph node dissection (2/6; 33.3%) steps.

Checcucci et al. conducted a prospective study involving 800 patients who underwent RARP to assess the impact of 3D models on PSM rate (22). They used high-definition 3D models based on multiparametric magnetic resonance imaging (mpMRI), with the 3D images available for consultation by the surgeon during the procedure or overlaid onto live images during surgery. Of the participants, 160 were enrolled in the 3D group, while the remaining 640 from a historical cohort served as the control group. The authors found that in the 3D group, the PSM rate was lower (25% vs. 35.1%, P=0.01), particularly among patients with suspected extracapsular extension on mpMRI.

Kwong et al. conducted a study on 2,468 patients who underwent RARP using an AI-based Side-specific Extra-Prostatic Extension Risk Assessment tool (SEPERA) (17). The machine learning model sequentially constructs decision trees, each reducing the misclassification error of previous trees. SEPERA predicted extraprostatic extension after RARP more effectively than other models, with no evidence of model bias and no significant difference in the area under the receiver operating characteristic curve when stratified by race, biopsy year, age, biopsy type, or biopsy location. The most common errors were false positives, particularly among older patients with high-risk diseases.

Ekşi et al. created a model to predict BCR post-RARP based on clinical parameters and machine learning (19). They analyzed a population of 368 patients with a BCR rate of 19.8%, reporting in multivariate analysis that neutrophil to lymphocyte ratio, PSA density, risk classification, PI-RADS score, T stage, presence or absence of PSM, and seminal vesicle invasion were predictive factors of BCR. The AUC was 0.915, with a sensitivity and specificity of 90.6% and 79.8%, respectively. They tried three different machine learning algorithms (i.e., Random Forest, K-nearest neighbor, and logistic regression) to predict BCR, analyzing 37 parameters and reporting AUC for receiver-operating characteristic curves of 0.95, 0.93, and 0.93, respectively.

Detection of the index lesion during surgery and surgical steps

Schiavina et al. evaluated the impact of an augmented reality 3D (AR-3D) model to guide nerve sparing during RARP (24). The authors included 26 patients, 23 with preoperative mpMRI-identified organ-confined index lesions and suspected extracapsular extension in the remaining ones. They reported that the initial, preoperative MR-based surgical plan for the nerve-sparing technique was modified due to the AR-3D model in 38.5% of patients. Notably, in 30% of these cases, surgery was adjusted towards a more radical approach, while in the remaining 70%, a less radical nerve-sparing approach was attempted. Furthermore, the intraoperative utilization of AR-3D technology (i.e., the intraoperative nerve-sparing plan) influenced the surgeon to modify their plan in 18 (34.6%) cases, with an overall appropriateness of 94.4%. Within these modifications, half the cases involved a shift towards a more radical approach (appropriateness 77.8%), while the other half transitioned to a less radical approach (appropriateness 88.9%). Finally, the AR-3D model exhibited promising accuracy in localizing the index lesion, achieving 70% sensitivity, 100% specificity, and 92% accuracy based on the 32-prostatic area map analysis. The authors support the feasibility of AR-3D guided surgery as a tool to enhance real-time identification of the index lesion and its potential to refine nerve-sparing approaches during RARP.

In support of these findings, Kaneko et al. reported on a proof concept study on the use of an algorithm that predicts the volume and location of clinically significant PCa (10). The authors used a convolutional neural network trained with the integration of multiparametric MR-ultrasound (US) image data and MRI-US fusion prostate biopsy trajectory-proven pathology data and compared it with independent radiologist reports. The concordance of the MRI index lesion with that in RARP specimens was significantly higher in the prediction model than in the radiologist’s reading (83% vs. 54%, P=0.036). In addition, clinically significant tumor volumes predicted with the model were more accurate than the radiologist’s reading (r=0.90, P<0.001).

Checcucci et al. assessed the precision of a 3D automated augmented reality system, which is AI-guided, in pinpointing the tumor’s position within the preserved neurovascular bundle after the excision phase during RARP (14). They prospectively include 34 men with cT1-3 PCa undergoing RARP. The virtual 3D model of the prostate enabled the identification of the tumor’s position within the preserved neurovascular bundle and facilitated a targeted excisional biopsy, preserving the remainder of the bundle. PSM rates were 0% in pT2 and 7.1% in pT3 patients. Using this augmented reality system, it was feasible to accurately locate the lesion on the neurovascular bundle in 87.5% of pT3 patients and executed a customized, 3D-guided nerve-sparing procedure even in cases of locally advanced diseases, all while maintaining oncological safety and minimizing PSM rates.

Evaluation of patients’ outcomes

A study conducted by Ma et al. included 80 nerve-sparing RARP procedures from two centers, to assess whether gesture types, the proportion of a gesture type, and expertise levels could predict the 1-year erectile function recovery (21). The authors found out that patients who recovered their erectile function had overall less hot cut (median 1.4% vs. 1.9%, P=0.016) but more peel/push (median 33.4% vs. 29.7%, P<0.001), in both the expert (median 275 cases) and super-expert group (median 3,000 cases). Two teams were instructed to build machine learning prediction models that considered both gesture sequences and clinical features. Both models achieved a moderately high ability to predict 1-year erectile function recovery when including either clinical features (i.e., Gleason score, age, body mass index, PSA, prostate volume) and surgical gestures or surgical gestures alone. These findings suggest that specific surgical gestures could be useful to achieve better surgical performances and functional outcomes after RARP. The models also highlighted a significant difference in surgical gestures and number of gestures when comparing the expert and super-expert subgroup, with the former using more gestures, more spread, more hook, and less coagulation, and the latter using fewer gestures, more cold-cut and coagulation but less peel/push and retraction. The same group also assessed the impact of patient and treatment variables, surgeon efficiency metrics, and surgeon technical skill scores, on models aimed at predicting urinary continence recovery defined as 0 or 1 safety pad/die (20). They found that APM that evaluate the most granular surgical movements, such as sub stitch maneuvers, and technical abilities seem to enhance the ability to predict urinary continence recovery following RARP. The assessment of suturing technical skills seems particularly crucial in foreseeing this critical patient outcome of functional recovery.

The first study to show that APM and machine learning algorithms may help assess surgical RARP performance and predict the postoperative length of stay was published in 2018 by Hung et al. (25). The Random Forest-50 algorithm demonstrated the highest efficacy, achieving an accuracy of 87.2% in predicting length of stay. Notably, the extended length of stay cases exhibited superior performance compared to the expected length of stay cases across various metrics: surgery duration (3.7 vs. 4.6 hours, P=0.007), length of stay (2 vs. 4 days, P=0.02), and Foley catheter duration (9 vs. 14 days, P=0.02). Additionally, patient outcomes forecasted by the algorithm showed significant correlations with actual outcomes in surgery duration (P<0.001, r=0.73), length of stay (P=0.05, r=0.52), and Foley catheter duration (P<0.001, r=0.45).

In 2019, the same group published a paper where the main purpose was to predict urinary continence recovery after RARP using a deep learning model (13). Robotic surgical APMs during RARP were reported for each step, and patient clinicopathological and continence data were captured prospectively from 100 contemporary procedures. They stratified eight surgeons based on the five top-ranked features (time-related metrics, instrument kinematic metrics, camera movement metrics, system event metrics, and wrist articulation metrics). The top four surgeons were categorized in Group 1/APMs. Continence was evaluated during each follow-up visit using the self-administered EPIC questionnaire. In this context, continence was defined as the absence of pads. The authors employed three datasets for predicting continence recovery: 16 clinicopathological features, 492 APMs, and a combined set comprising 16 clinicopathological features and 492 APMs, totaling 508 measures. Out of the RARP cases considered in this study, continence was achieved in 79 patients (79%), with a median time to urinary continence of 126 days. Three of the top-ranked features were APM measured during the vesicourethral anastomosis, and one was measured during the prostatic apical dissection. APM were ranked higher by the model than clinicopathological features. Patients in Group 1/APMs had superior rates of urinary continence at 3 and 6 months postoperatively (47.5% vs. 36.7%, P=0.034, and 68.3% vs. 59.2%, P=0.047, respectively).

Evaluation of surgical performances

Khanna et al. developed and trained an algorithm using 292 RARP videos that were manually annotated with the corresponding surgical steps (15). Subsequently, the algorithm’s performance was evaluated on an independent set of 182 RARP procedures. The analysis revealed a high concordance (92.8%) between the AI-based automated video analysis and manual human video annotation. Notably, the algorithm achieved the highest accuracy for the vesicourethral anastomosis step (97.3%), while the final inspection and extraction step showed the lowest accuracy (76.8%). These findings suggest that automated surgical video analysis using this approach has promising applications in various domains, including surgeon video review, surgical training and education, and quality and safety benchmarking.

Chen et al. focused on the application of APM during vesicourethral anastomosis to distinguish surgeon experience (23). The study reported APM during the vesicourethral anastomosis for each overall stitch (Ctotal) and its sub-stitch components: needle handling/targeting, needle driving, and suture cinching during anastomosis. The data were then recorded into a systems data recorder, organized into three datasets [GlobalSet (whole stitch), RowSet (independent sub-stitches), and ColumnSet (associated sub-stitches)] and applied to three machine learning models (AdaBoost, Gradient boosting, and Random Forest). Participant surgeons were divided into novices, ordinary experts, and super-experts based on the number of cases they worked on. When it came to differentiating between novice and expert surgeons, ColumnSet, providing sub-stitch details along with association to specific stitches, performed with the highest accuracy, followed by RowSet and the worst being GlobalSet—the only one that did not account for sub-stitch—and the overall best combination of dataset and model was ColumnSet by Random Forest model. Furthermore, the study showed that the algorithms were more accurate in differentiating between super experts and ordinary experts than distinguishing novices versus experts. When comparing experts to novices, APM showed the most differences in total path length of non-dominant instruments, articulation of both dominant and non-dominant instruments, and total moving time of the camera, whereas features during needle handling/targeting phase of suturing turned out to be the most important differences between super experts and ordinary experts, especially mean path length of camera movements, total path length of dominant instrument and articulation of dominant instrument. Nonetheless, the study has its limits, most importantly being based on a single center’s experience with surgeons who may share a similar surgical technique.

Zuluaga et al. described a surgical AI platform system that generated real-time annotations and identified operative safety milestones during live RARP (12). This was realized through trained algorithms, conventional video recognition, and novel video transfer network technology which can capture clips in full context, allowing automatic recognition and surgical mapping in real time.

Another helpful tool based on AI was presented by Checcuci et al. who, using convolutional neural networks, developed a system capable of forecasting instances of intraoperative bleeding (18). The authors found that the system was able to alert surgeons to potential bleeding incidents with an impressive event recognition accuracy of 90.63%.

Discussion

Key findings

In this scoping review, we reported the current knowledge regarding the use of AI in RARP. This covers several aspects from surgical planning to intraoperative decision-making, evaluation of surgical performance, and assessment of patients’ postoperative outcomes.

Oncological radicality is certainly one of the most important objectives of the surgical treatment of PCa. Predicting that some patients may have a higher probability of PSM can help the surgeon prevent such an outcome. The model probably most easily applicable in daily clinical practice is the one proposed by Hao et al. (26). Their nomogram is based on clinical parameters and is easy to use in daily clinical practice, also thanks to the online calculator they have developed. However, this system overlooks the variances among individual surgeons, which instead are considered by Lee et al. through the APM (11). The model they developed can help each surgeon predict the probability of PSM based on the technique used.

New technologies play a fundamental role in enhancing surgical technique as well. The 3D models analyzed by Checcucci et al. enable greater precision during RARP in terms of oncological radicality, significantly reducing the PSM rate, choosing a nerve-sparing approach in the best candidate, anticipating bleeding and thereby allowing for an improved outcome for the patient (14,18,22).

The model developed by Kwong et al., aiming to predict extraprostatic extension, can also be very useful for surgical planning, especially concerning the applicability of nerve-sparing techniques to maintain oncological radicality (17). Predicting post-RARP BCR can also be a significant advantage in terms of oncological radicality. The model proposed by Ekşi et al. can help urologists understand which patients are at higher risk of BCR and place them under closer follow-up, improving the final oncological outcome (19).

Urinary continence is another pivotal outcome after RARP. Factors affecting postoperative urinary continence recovery have been extensively investigated, including clinicopathological characteristics and nerve-sparing and reconstruction techniques. Surgeons who perform RARP continuously accumulate experience and develop their technical skills, resulting in improved urinary continence and sexual health. To date, the ‘gold standard’ method of weighing surgeon performance has been through determining prior surgical experience (caseload) or manual surgical evaluation by peer surgeons (27). However, evaluating surgical performance by caseload alone may be inaccurate or inconsistent as it often is self-reported by the surgeon. A fledgling alternative to manual assessment is APM. These instrument motion tracking and events metrics are derived from computer-based data recording devices. Recent developments involve the application of APM in conjunction with machine-learning algorithms. These tools have been shown to predict effectively peri-operative and short-term clinical outcomes following RARP (23,25).

Automated surgical video analysis using this approach has promising applications in various domains, including surgeon video review, surgical training, education, quality and safety benchmarking.

The findings of the Khanna et al. study demonstrate high concordance (92.8%) between AI-based automated surgical video analysis and manual annotation, highlighting the potential of this technology to revolutionize various aspects of RARP (15). The ability to accurately identify and segment surgical steps within RARP videos unlocks a range of promising medical applications.

Automatic identification and segmentation of key surgical steps can help urologists focus on specific aspects of the procedure, reducing the overall review time of surgical videos. This can be particularly beneficial for training purposes or complex case reviews. Surgical trainees can utilize this technology to review in detail specific steps of RARP procedures, enhancing their understanding of the surgical workflow (12,15). Additionally, the ability to objectively assess trainee performance by analyzing their surgical videos can provide valuable feedback and identify areas for improvement. Training can be also improved by using machine learning models which highlight a significant difference in surgical gestures and number of gestures when comparing the expert and super-expert surgeons, with the former using more gestures, more spread, more hook, and less coagulation (21). These results suggest that not only the types of gestures but also likely the execution and context of gestures matter for outcomes and indeed this can be very helpful to improve outcomes of novice surgeons.

Strengths and limitations

In the context of RARP, AI has shown promising applications in planning and enhancing surgical outcomes. The strengths of AI algorithms rely on the analysis of patient data, including MRI, to assist in preoperative planning. This helps surgeons identify the optimal approach and plan the surgery with greater precision. AI can also provide decision support by analyzing complex datasets and offering recommendations based on historical patient data, outcomes, and best practices. This assists surgeons in making informed decisions during the surgical planning phase. A further advantage of AI-powered image analysis is helping the surgeon in real-time navigation during the procedure. This includes identifying critical structures, guiding the surgeon to target areas, and providing continuous feedback to ensure accurate and safe procedures. Machine learning models can analyze patient data to predict the likelihood of postoperative complications. This allows healthcare professionals to take preventive measures and optimize postoperative care. Moreover, AI can analyze data from a large number of surgeries to identify patterns, trends, and best practices. This continuous learning process can contribute to refining surgical techniques and improving overall outcomes over time. Finally, AI technologies enable remote collaboration and assistance during surgeries using current high-speed connections. Surgeons can receive real-time guidance and support from experts, regardless of geographical location, improving the overall quality of care.

Limitations of AI application for RARP are mainly related to its poor availability and costs, particularly in resource-constrained healthcare settings. In addition, implementing AI technologies in surgical settings requires significant investment in infrastructure, training, and maintenance. For some healthcare institutions, especially those with limited resources, the upfront costs associated with adopting AI may outweigh the potential benefits.

We should also consider that AI systems are typically trained on specific datasets and may struggle to adapt to variations in patient anatomy or unexpected surgical challenges during the procedure. Finally, integrating AI technologies into existing robotic surgical platforms may pose technical challenges and require significant modifications to the workflow and infrastructure of surgical departments.

Finally, our review relies on only 17 studies with different aspects of the application of AI based on still modest data and future studies are demanding to confirm this preliminary data.

Implications and actions needed

There are ethical and legal considerations surrounding the use of AI in surgery, including issues related to liability, patient consent, and privacy. Ensuring compliance with regulatory standards and addressing potential concerns about algorithmic bias and transparency is essential for its future application in daily practice.

Conclusions

The integration of AI in RARP represents a transformative advancement in surgical practice. Through machine learning algorithms, AI augments surgical precision, enhances decision-making processes and facilitates personalized patient care. This holds immense potential to improve surgical outcomes, reduce operative times, improve teaching, and mitigate complications, ultimately advancing the standard of care for patients undergoing RARP. However, this should be balanced against the current costs of implementation of robotic platforms with such a technology.

Acknowledgments

Funding: None.

Footnote

Provenance and Peer Review: This article was commissioned by the editorial office, Chinese Clinical Oncology for the series “New Evidence and Advances in Surgical Treatment of Prostate Cancer”. The article has undergone external peer review.

Reporting Checklist: The authors have completed the PRISMA-ScR reporting checklist. Available at https://cco.amegroups.com/article/view/10.21037/cco-24-52/rc

Peer Review File: Available at https://cco.amegroups.com/article/view/10.21037/cco-24-52/prf

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://cco.amegroups.com/article/view/10.21037/cco-24-52/coif). The series “New Evidence and Advances in Surgical Treatment of Prostate Cancer” was commissioned by the editorial office without any funding or sponsorship. D.C. served as the unpaid Guest Editor of the series and serves as an unpaid editorial board member of Chinese Clinical Oncology from August 2022 to July 2024. S.F. also served as the unpaid Guest Editor of the series. The authors have no other conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Cancer Today. Globocan. 2024 [cited 2024 Apr 4]. Available online: https://gco.iarc.fr/today/en/dataviz/bars?mode=cancer&group_populations=1&key=total&sexes=1
2. Siegel RL, Giaquinto AN, Jemal A. Cancer statistics, 2024. CA Cancer J Clin 2024;74:12-49. [Crossref] [PubMed]
3. Mottet N, van den Bergh RCN, Briers E, et al. EAU-EANM-ESTRO-ESUR-SIOG Guidelines on Prostate Cancer-2020 Update. Part 1: Screening, Diagnosis, and Local Treatment with Curative Intent. Eur Urol 2021;79:243-62. [Crossref] [PubMed]
4. Adhikari K, Penmetsa GK, Krishnappa D, et al. Revolutionizing urology: the advancements and applications of robotic platforms. J Robot Surg 2024;18:106. [Crossref] [PubMed]
5. Ilic D, Evans SM, Allan CA, et al. Laparoscopic and robotic-assisted versus open radical prostatectomy for the treatment of localised prostate cancer. Cochrane Database Syst Rev 2017;9:CD009625. [Crossref] [PubMed]
6. Galfano A, Bocciardi AM. Retzius-sparing RARP: improving functional outcomes. Nat Rev Urol 2021;18:73-4. [Crossref] [PubMed]
7. Rong G, Mendez A, Bou Assi E, et al. Artificial Intelligence in Healthcare: Review and Prediction Case Studies. Engineering 2020;6:291-301. [Crossref]
8. Hashimoto DA, Rosman G, Rus D, et al. Artificial Intelligence in Surgery: Promises and Perils. Ann Surg 2018;268:70-6. [Crossref] [PubMed]
9. Tricco AC, Lillie E, Zarin W, et al. PRISMA Extension for Scoping Reviews (PRISMA-ScR): Checklist and Explanation. Ann Intern Med 2018;169:467-73. [Crossref] [PubMed]
10. Kaneko M, Fukuda N, Nagano H, et al. Artificial intelligence trained with integration of multiparametric MR-US imaging data and fusion biopsy trajectory-proven pathology data for 3D prediction of prostate cancer: A proof-of-concept study. Prostate 2022;82:793-803. [Crossref] [PubMed]
11. Lee RS, Ma R, Pham S, et al. Machine Learning to Delineate Surgeon and Clinical Factors That Anticipate Positive Surgical Margins After Robot-Assisted Radical Prostatectomy. J Endourol 2022;36:1192-8. [Crossref] [PubMed]
12. Zuluaga L, Rich JM, Gupta R, et al. AI-powered real-time annotations during urologic surgery: The future of training and quality metrics. Urol Oncol 2024;42:57-66. [Crossref] [PubMed]
13. Hung AJ, Chen J, Ghodoussipour S, et al. A deep-learning model using automated performance metrics and clinical features to predict urinary continence recovery after robot-assisted radical prostatectomy. BJU Int 2019;124:487-95. [Crossref] [PubMed]
14. Checcucci E, Piana A, Volpi G, et al. Three-dimensional automatic artificial intelligence driven augmented-reality selective biopsy during nerve-sparing robot-assisted radical prostatectomy: A feasibility and accuracy study. Asian J Urol 2023;10:407-15. [Crossref] [PubMed]
15. Khanna A, Antolin A, Bar O, et al. Automated Identification of Key Steps in Robotic-Assisted Radical Prostatectomy Using Artificial Intelligence. J Urol 2024;211:575-84. [Crossref] [PubMed]
16. Nakamura W, Sumitomo M, Zennami K, et al. Combination of deep learning and ensemble machine learning using intraoperative video images strongly predicts recovery of urinary continence after robot-assisted radical prostatectomy. Cancer Rep (Hoboken) 2023;6:e1861. [Crossref] [PubMed]
17. Kwong JCC, Khondker A, Meng E, et al. Development, multi-institutional external validation, and algorithmic audit of an artificial intelligence-based Side-specific Extra-Prostatic Extension Risk Assessment tool (SEPERA) for patients undergoing radical prostatectomy: a retrospective cohort study. Lancet Digit Health 2023;5:e435-45. [Crossref] [PubMed]
18. Checcucci E, Piazzolla P, Marullo G, et al. Development of Bleeding Artificial Intelligence Detector (BLAIR) System for Robotic Radical Prostatectomy. J Clin Med 2023;12:7355. [Crossref] [PubMed]
19. Ekşi M, Evren İ, Akkaş F, et al. Machine learning algorithms can more efficiently predict biochemical recurrence after robot-assisted radical prostatectomy. Prostate 2021;81:913-20. [Crossref] [PubMed]
20. Trinh L, Mingo S, Vanstrum EB, et al. Survival Analysis Using Surgeon Skill Metrics and Patient Factors to Predict Urinary Continence Recovery After Robot-assisted Radical Prostatectomy. Eur Urol Focus 2022;8:623-30. [Crossref] [PubMed]
21. Ma R, Ramaswamy A, Xu J, et al. Surgical gestures as a method to quantify surgical performance and predict patient outcomes. NPJ Digit Med 2022;5:187. [Crossref] [PubMed]
22. Checcucci E, Pecoraro A, Amparore D, et al. The impact of 3D models on positive surgical margins after robot-assisted radical prostatectomy. World J Urol 2022;40:2221-9. [Crossref] [PubMed]
23. Chen AB, Liang S, Nguyen JH, et al. Machine learning analyses of automated performance metrics during granular sub-stitch phases predict surgeon experience. Surgery 2021;169:1245-9. [Crossref] [PubMed]
24. Schiavina R, Bianchi L, Lodi S, et al. Real-time Augmented Reality Three-dimensional Guided Robotic Radical Prostatectomy: Preliminary Experience and Evaluation of the Impact on Surgical Planning. Eur Urol Focus 2021;7:1260-7. [Crossref] [PubMed]
25. Hung AJ, Chen J, Che Z, et al. Utilizing Machine Learning and Automated Performance Metrics to Evaluate Robot-Assisted Radical Prostatectomy Performance and Predict Outcomes. J Endourol 2018;32:438-44. [Crossref] [PubMed]
26. Hao Y, Zhang Q, Hang J, et al. Development of a Prediction Model for Positive Surgical Margin in Robot-Assisted Laparoscopic Radical Prostatectomy. Curr Oncol 2022;29:9560-71. [Crossref] [PubMed]
27. Hajiha M, Baldwin DD. Factors affecting urinary incontinence during robotic radical prostatectomy. Transl Androl Urol 2018;7:S93-5. [Crossref] [PubMed]