OA02.7. CT-Based Segmentation AI and 3D Computer Graphics for Esophageal Cancer Surgery

Abstract

Esophageal Cancer: New Diagnostic Modalities (Including PET, PET-CT, New Tracers) Accurate understanding of complex mediastinal anatomy is essential for safe esophageal cancer surgery. However, CT-based segmentation and 3D reconstruction of the esophagus and mediastinal lymph nodes remain difficult because of low contrast and unclear anatomical boundaries. This study included 180 esophageal cancer patients. One-millimeter contrast-enhanced CT images were manually annotated to train a CT-based segmentation AI for the esophagus and mediastinal lymph nodes with an industry partner. Segmentation outputs were integrated into a 3DCG visualization system developed with another partner (Viewtify®). A split-bolus CT protocol enabled simultaneous arterial and venous enhancement in a single acquisition. Probability-based smoothing was applied during 3D reconstruction, and the resulting models were used for preoperative simulation and navigation assessment. The segmentation AI achieved good performance, with a Dice coefficient of 0.88 for both the esophagus and mediastinal lymph nodes. The split-bolus CT protocol provided stable, same-phase enhancement of vessels and organs, improving spatial consistency for 3D reconstruction. Integrated 3D visualization allowed intuitive understanding of spatial relationships among the esophagus, lymph nodes, vessels, and airway, supporting preoperative planning and intraoperative navigation. Combining split-bolus CT with automated segmentation AI and 3D visualization enables a practical framework for esophageal cancer surgery, supporting standardized preoperative assessment, intraoperative decision-making, and surgical education.