Deep learning-based vessel and nerve recognition model for lateral lymph node dissection: a retrospective feasibility study

Le résumé fourni par la source

PURPOSE: Lateral lymph node dissection for rectal cancer is challenging because of the presence of blood vessels and nerves essential for postoperative genitourinary function and leg movements. Identifying these structures during surgery is crucial. We developed a deep learning-based semantic segmentation model to recognize and visualize critical anatomical structures during laparoscopic lateral lymph node dissection automatically. METHODS: Intraoperative video data from laparoscopic lateral lymph node dissections performed on 22 patients between 2018 and 2021 were used. Specific scenes from the beginning to end of the procedures were extracted and divided into still images, which were annotated to delineate the external iliac artery, external iliac vein, and obturator nerve. The model was trained with pixel-level annotation labels, and its performance was evaluated using precision, recall, and the Dice coefficient through five-fold cross-validation. RESULTS: Overall, 992 images were extracted from 22 lateral lymph node dissection videos. The Dice coefficient values were 0.789 (± 0.009), 0.736 (± 0.033), and 0.574 (± 0.082) for the obturator nerve, external iliac artery, and external iliac vein, respectively. The model's inference speed was 12.7 fps, corresponding to processing one still image in 0.08 s, enabling near real-time intraoperative analysis. CONCLUSION: The deep learning-based semantic segmentation model automatically recognized the obturator nerve, external iliac artery, and external iliac vein during laparoscopic lateral lymph node dissection, achieving reasonable segmentation accuracy as measured by the Dice coefficient. This technology will be used as a foundation for developing surgical navigation systems to improve the safety and efficiency of lateral lymph node dissection procedures.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.