Agreement and precision of an AI-based measurement tool during robotic-assisted sleeve gastrectomy

Size and shape of the stomach following a sleeve gastrectomy (SG) appear to impact operative outcomes; however, standardization of these variables is difficult without accurate measurements. Manual measurement techniques are prone to inter-operator variability and can be cumbersome. Artificial Intelligence (AI)-based measurement tools integrated in robotic platforms may provide real-time, precise anatomical assessment. The purpose of this study is to evaluate the preliminary agreement of an AI-based measurement tool relative to conventional ruler measurement during robotic-assisted sleeve gastrectomy (SG) surgery, the most common type of bariatric surgery performed worldwide.

Adult patients (n = 12, > 18 years) undergoing robotic-assisted SG between August 2025 and January 2026 were included. Measurements (cm) at the widest point of the gastroesophageal (GE) junction and the narrowest point at the incisura were obtained using both point-to-point (P2P) and contour approaches. Data were summarized using mean ± standard deviation (SD). Agreement was assessed using Pearson's correlation analysis with 95% confidence intervals, partial correlation adjusting for BMI, and Bland-Altman analysis.

All 12 patients were female with a mean BMI of 46.3 ± 6.5 kg/m². Pearson correlation coefficients between AI and ruler measurements ranged from 0.79 to 0.90 across all measurement types. P2P measurements at the GE junction demonstrated the strongest correlation (ρ = 0.90; 95% CI: 0.66-0.97), while contour measurements at the incisura showed the lowest (ρ = 0.79; 95% CI: 0.39-0.94). Bland-Altman analysis demonstrated low bias across all measurements (absolute mean differences < 0.1 cm) with no evidence of proportional bias. Partial correlation analysis adjusting for BMI showed consistent results.

This pilot study demonstrates preliminary agreement between commercially available robotic-integrated AI measurement tool and standard ruler measurement during live SG, particularly for P2P measurements. These hypothesis-generating findings support the potential of AI-assisted measurement to enhance surgical precision and efficiency, providing a foundation for future larger-scale, multicenter validation studies.