FP5.3 Development of a Deep Learning Model for Ultrasound-Guided Arthroscopic Surgery in Femoroacetabular Impingement: A Pilot Study

Abstract

Abstract Background Arthroscopic surgery for femoroacetabular impingement (FAI) is effective, and technological advancements continue to improve surgical outcomes. One such improvement is using CT navigation for precise osteochondroplasty, especially for cam corrections. Concurrently, ultrasound's role in diagnosis, treatment, and ultrasound-guided surgery is gaining attention across medical fields. Therefore, we are working to integrate machine learning-based image recognition to enhance ultrasound-guided hip arthroscopic surgery, aiming for greater accuracy and efficiency. Objective This pilot study aimed to train a deep learning model to recognize the tip of an ablator bar in intraoperative ultrasound videos during osteochondroplasty and evaluate its real-time accuracy. The goal was to develop a tool that could support ultrasound-guided arthroscopic surgery. Methods The study included 27 hips diagnosed with FAI and treated with arthroscopic osteochondroplasty between 2023 and 2024. Intraoperative ultrasound videos were captured during osteochondroplasty using an ablator bar. These videos were analyzed using a deep learning algorithm (YOLO v8), with the ablator bar’s tip labeled in rectangular regions. To evaluate the model’s performance, two validation methods were used: Verification A (splitting the 19,948 frames into 9,974 training and 9,974 test frames) and Verification B (splitting the 27 videos into 14 for training and 13 for testing). The agreement rate was assessed using mean Average Precision (mAP50-95). Results In Verification A, the agreement rate between the model’s predictions and the actual tip location was 99.4%. In Verification B, the agreement rate dropped to 70.6%. This decrease suggests that more data and refinement are needed to improve the model’s generalizability. Discussion Achieving high-accuracy detection of the ablator bar in ultrasound images requires further case accumulation and model refinement. While X-ray remains essential for precise cam corrections, integrating ultrasound-guided arthroscopic surgery could reduce reliance on radiation exposure and pre-operative CT imaging, offering significant clinical benefits in terms of safety and surgical efficiency.

Ähnliche Arbeiten