A novel image-guided laparoscopic liver resection with integrated fluorescent imaging and artificial intelligence: A preliminary study.

Abstract

568 Background: Technology innovations improve surgical techniques. ICG fluorescence imaging is beneficial for real time navigation tool during surgery and enables accurate surgery. Recently, the prediction of anatomical structures by artificial intelligence (AI) is expected to support surgeons as navigation tool. We developed a visual-support AI system to perform the laparoscopic liver resection (LLR) using ICG fluorescence imaging and AI technology. Methods: Procedure: ICG fluorescence imaging was used for liver mapping and securing surgical margin. AI model development and evaluation: Over 430 images extracted from 13 videos of LLR for hepatocellular carcinoma and colorectal liver metastasis was used to develop and evaluate the AI algorithm at Anaut Inc. The annotation was performed on video frames capturing a LLR performed by 2 surgeons for hepatic vein and Glisson sheath. The performance of AI on unlearned LH videos was evaluated quantitatively using Intersection-Over-Union (IOU) and Dice coefficients, and qualitatively on a five-point scale (from 1 to 5) by 10 hepato-biliary-pancreatic surgeons using 35 randomly extracted still images from the analyzed videos. Results: ICG fluorescence imaging could visualize the demarcation line and clarified the boundaries of the liver transected surface. The hepatic veins and Glisson sheath were appeared in the parenchymal transection plane during segmentectomy and securing surgical margin, and this AI system enabled to recognize and display these vascular structures through color-coding under bleeding and ICG fluorescent imaging without visual discrepancies. The AI model accurately recognized vascular structures of any size (IoU=0.33, Dice=0.44) in real-time (<0.12 s) In the qualitative evaluation by surgeons, the mean sensitivity score was 4.10 ± 0.92 (range 3.61–4.76). Additionally, this AI system also recognized vascular structures during robot-assisted liver resection. Conclusions: This is the first study demonstrating that AI can precisely identify vascular structures during LLR. Novel image-guided LLR using ICG fluorescent imaging and artificial intelligence is expected to provide safer and more accurate LLR, thereby contributing to reduce complications and improve survival rates.