Landmark display system for laparoscopic inguinal hernia repair using artificial intelligence

Abstract

Background: Chronic postoperative inguinal pain (CPIP) is a major complication of inguinal hernia repair and significantly affects patients’ quality of life. Despite the widespread use of transabdominal preperitoneal repair (TAPP), CPIP still occurs. In TAPP, nerves are not directly exposed, and surgeons avoid complications by identifying the trapezoid of disaster (ToD) using anatomical landmarks. This study evaluates an AI-based system that displays these landmarks on intraoperative endoscopic images. Methods: We analyzed 62 randomly selected TAPP cases (73 hernias) from 188 procedures (222 hernias) performed at our institution. A total of 3323 images of the myopectineal orifice were labeled for training. We developed a Feature Pyramid Network (FPN) segmentation model using EfficientNetB7. The model was tested on 10 new cases to identify three key landmarks: vas deferens, gonadal vessels, and inferior epigastric vessels. Expert surgeons validated the prototype system. Results: Postoperative pain was observed in 5.4% (12 cases) for Numerical rating scale (NRS) ≥ 1 and 2.2% (5 cases) for CPIP (NRS ≥ 3). The model achieved Dice coefficients of 0.67 (vas deferens), 0.68 (gonadal vessels) and 0.70 (inferior epigastric vessels). Anatomical landmarks were successfully displayed on surgical images. Expert evaluation confirmed correct recognition in 90% of cases. Conclusions: The AI model accurately identified key anatomical landmarks in TAPP and demonstrated anatomical validity. Surgical safety may be improved by avoiding nerve damage through visualization of key structures.