MCPD-YOLOv3: A Novel Lightweight Detection Model for Surgical Instruments in Laparoscopic Images

Laparoscopic surgical instruments detection is necessary in computer-aided minimally invasive surgery. Most current methods suffer from unsatisfied performance and low detection speed.

In this paper, a framework called MCPD-YOLOv3 is proposed to balance the efficiency and effectiveness of laparoscopic surgical instruments detection. It effectively fuses feature maps using a parallel manner, and adopts various lightweight strategies to design a lightweight model. Besides, DIoU is employed to improve the recall performance.

The proposed method achieved the mAP of 99.47% and 97.65% at 49.81 FPS for the ATLAS Dione and m2cai16-tool-locations datasets, respectively, with a compact model size of 12.4M and a low FLOPs count of 7.44G.

These results highlight that MCPD-YOLOv3 excels in high detection performance and rapid response. The model's efficiency in parameter size and FLOPs demonstrates its suitability for applications requiring rapid processing and precise detection, making it a valuable tool for real-time surgical instrument detection in challenging environments.