Abstract
<title>Abstract</title> <p>Background Robotic surgery, using systems like the da Vinci system, has transformed minimally invasive surgery with its enhanced mechanical dexterity and high-definition visualization. However, cognitive skills still constitute a procedural hurdle when it comes to complex interventions. Artificial Intelligence (AI) serves as a cognitive add-on to optimize decision-making in surgery. This study critically evaluates the clinical implications of integrating AI into surgical precision and postoperative morbidity, while specifically addressing common challenges in the literature that include the transparency of the algorithms and standardisation of definitions for surgical outcomes. Methods This was a retrospective study, carried out on a single-center cohort of 180 adult patients, who underwent robotic-assisted surgery from January 2020 to December 2024. To create a homogeneous cohort, bariatric procedures were excluded and patients were evenly distributed between 3 specialties: radical prostatectomy, colorectal resection and hysterectomy. Patients were split into two groups: AI enhanced robotic surgery arm (n = 90) and conventional control arm (n = 90). A pre-validated 3D Convolutional Neural Network (CNN) based on a ResNet-50 architecture was embedded into the da Vinci Xi workflow for real-time tissue segmentation, motion scaling and prediction of complications. Complications were evaluated according to strict definitions (Clavien-Dindo classification) and pathological margin evaluation was standardized. Findings: Intraoperative efficiency and the use of resources were significantly improved for the AI-enhanced robotic cohort, with an 18% decrease in mean operative time (162 ± 22.1 vs. 198 ± 30.4 min, p < 0.001) and a 21% decrease in intraoperative blood loss (p = 0.03). Analysis of the postoperative outcome showed that major morbidities were significantly reduced in the AI group, with the reduction of surgical site infections (8.2% vs. 14.7%, p = 0.02) and the mean hospital stay reduction of 1.7 days, which boosts the cost-effectiveness of the hospital. Additionally, the Area Under the Curve (AUC) value of the CNN predictive model was very high in predicting surgical errors and complications, with a value of 0.92. The 12-month tumor recurrence rates were significantly lower in the AI cohort (2.1% vs. 5.4%, p = 0.02) in the long-term follow-up. Conclusion CNNs in robotic surgery introduce AI models that improve robotic platforms by enabling them to adapt to patients' cognitive needs. This synergy helps to greatly reduce intra-operative error, post-op complications, and the use of healthcare resources. Further studies are needed to achieve this paradigm shift: future research must focus on frameworks for Explainable AI (XAI) and multi-center testing to ensure fair and safe clinical deployment.</p>