Surgical Tool Detection on CholecTrack20 Using Lightweight Deep Learning Models

Abstract

With the rapid development of minimally invasive surgery (MIS) and robot-assisted surgery, real-time, accurate, and robust surgical instrument detection and tracking has become a core research focus in medical AI. This review summarizes the current state of surgical tool detection and tracking, with a particular focus on the application of lightweight deep learning models on the CholecTrack20 dataset. Models such as MobileNetV2 and YOLOv8n demonstrate promising deployment potential and performance on embedded and low-computation platforms. We analyze their advantages and limitations in small-object detection, occlusion handling, multi-view tracking, and end-to-end real-time inference, and discuss potential improvements through multi-frame fusion, feature pyramid networks, lightweight attention modules, and data augmentation strategies. Furthermore, future research directions are outlined, including multimodal perception (vision + tactile/force feedback), explainable AI (XAI), and uncertainty estimation to ensure clinical safety and regulatory compliance. Overall, lightweight models offer practical deployment value for surgical tool detection and tracking, and provide a feasible pathway toward intelligent, multimodal surgical systems.

Ähnliche Arbeiten