Automated Patient-Specific Pneumoperitoneum Model Reconstruction for Surgical Navigation Systems in Distal Gastrectomy

Abstract

The development of surgical navigation systems is critical in computer-aided surgery for achieving precision surgery. This study introduces an innovative framework for reconstructing patient-specific pneumoperitoneum models in Minimally Invasive Gastrointestinal Surgery (MIGS) navigation systems. Leveraging preoperative CT images and intraoperative 3D scan landmark data from 210 gastric cancer patients, we propose a data-driven approach to pneumoperitoneum reconstruction. Unlike conventional physics-based methods, our framework utilizes landmark displacement regression models to capture patient-specific deformation information. Furthermore, we utilize a CNN-based model to extract fat and muscle area from CT images and train the displacement regression model to reflect patient-specific characteristics. The efficacy of our approach is evaluated through comprehensive evaluation using traditional statistical and neural networks-based methods approaches. The code for reproducibility is available at github.com/PRIME-MICCAI24-APSP/APSP.