Development and validation of 3D-printed anatomically accurate liver model for endoscopic ultrasound-based endo-hepatology training (with video).

BACKGROUND: Ex vivo endoscopic ultrasound (EUS) training relies on animal tissue, which is anatomically variable, perishable, and requires dedicated endoscopes. As EUS-based liver assessment (EUS-LA)-including parenchymal imaging, shear-wave elastography (SWE), attenuation imaging, and EUS-guided liver biopsy (EUS-LB)-expands, an anatomically accurate, tissue-free training platform could be a valuable educational tool. We developed a synthetic training tool that integrated a gelatin liver and silicone foregut and assessed its validity. METHODS: CT data from a healthy female were extracted to three-dimensionally print a liver replica. The replica was used to create a negative mold to cast a gelatin matrix around color-coded vasculature, then integrated with silicone esophagus/stomach/duodenum model to recreate normal foregut/liver anatomy. Sixteen endoscopists (12 experts from five centers in two countries; 4 novices) performed standardized EUS-LA across four tasks. Content validity (12 experts) was summarized with a content-validity index (CVI); inter-rater agreement used Gwet's AC1 with 95% confidence intervals (CI); face validity and global satisfaction were assessed in all 16. RESULTS: Composite CVI was 0.85 (AC1 0.68, 95% CI 0.54-0.83; p < 0.0001). CVI realism was 0.87, with perfect endorsement for liver parenchyma (1.00; AC1 0.90) and high endorsement for quantitative imaging (0.95; AC1 0.88); vascular realism was lower (0.73), with expert agreement no better than chance (AC1 0.41; p = 0.17). CVI relevance was 0.90 and representativeness 0.77. The complete workflow, including the EUS-LB maneuver, vessel avoidance, and SWE/attenuation acquisition, was completed by 100% of evaluable participants. All endorsed use before ex vivo training, and 93.8% would recommend the model; global satisfaction was 4.25 (IQR 4-5). CONCLUSIONS: The anatomically accurate, synthetic training model showed strong content and face validity and supported the complete diagnostic endo-hepatology workflow in a reproducible manner. Vascular realism is the principal target for refinement.