Gastric volume reduction estimation following Endoscopic Sleeve Gastroplasty through Image-based Analysis

Endoscopic Sleeve Gastroplasty (ESG) is proven to induce sustained weight loss and improvement of obesity-related comorbidities. ESG is performed using endoscopic full-thickness sutures leading to gastric restriction and shortening. While several studies have described technical variations in ESG and their potential impact on outcomes, the effect of ESG on gastric volume reduction has not yet been quantified. This study aims to objectively measure stomach volume reduction before and after ESG using image-based assessment.

Videos of ESG procedures were collected at a single institution. The procedure was subdivided into phases, and matching frames of the "preoperative" and "end" phase were selected for analysis by an expert ESG endoscopist. DepthAnythingv2 was used to generate inverse depth images for the selected frames. Within the stomach, a monotonic rescaling of inverse depth was applied; the per-pixel relative change (pre-post)/pre was computed and used as apparent volumetric reduction metric.

ESG videos performed by a single operator across its learning curve were annotated. Eighteen videos with matching frames were selected. Across the early and late learning curve groups, the estimated volume change went from a mean of 22.8% (SD 16.14%), up to 37.64% (SD 12.05%), with statistically significant % volume change (p = 0.080) when analysing the ordered trend across all three learning curve point. The patients achieved a mean Total Weight Loss% of 13.1% (SD 5.5%) and 9.6% (SD 8.5%) at 6 and 12 months, respectively, whereas Excess Weight Loss% was 33.3% (SD 23.2%) and 26% (SD 22.2%) for the same timepoints.

Using depth images generated from ESG videos, gastric volume reduction pre- and post-ESG was estimated, demonstrating greater volume reduction along the operator learning curve, and confirming the feasibility and utility of this methodology. Integrating image-base volume estimation through computer vision real-time systems could assist proficiency assessment and operative notes generation.