DBH-YOLO: a surgical instrument detection method based on feature separation in laparoscopic surgery

Area of Science:

• Medical Imaging
• Computer Vision
• Surgical Technology

Background:

• Accurate analysis of surgical instruments in laparoscopic videos is crucial for postoperative quality assessment.
• Improved detection aids in providing patients with scientific solutions for surgical complication healing.

Purpose of the Study:

• To propose an end-to-end algorithm for accurate detection of surgical instruments in laparoscopic videos.
• To address challenges in both localization and classification of surgical instruments.

Main Methods:

• Introduction of Dual-Branched Head (DBH) and Overall Intersection over Union Loss (OIoU Loss) for enhanced detection.
• Development of DBHYOLO, an effective method for laparoscopic surgery detection in complex scenarios.
• Manual annotation of a new dataset, LGIL, for laparoscopic gastric cancer resection surgical instrument localization.

Main Results:

• DBHYOLO achieved high mean Average Precision (mAP) values: 96.8% (m2cai16-tool-locations), 95.6% (LGIL), and 98.4% (Onyeogulu).
• The proposed method outperformed classical models in accuracy and reduced missed detection cases.
• The model effectively distinguishes between surgical instrument classes with high visual similarity.

Conclusions:

• The DBH-YOLO method effectively addresses inaccurate surgical instrument detection from both classification and localization perspectives.
• Experimental results on three datasets validate the performance and generalization capability of DBH-YOLO.
• The proposed approach offers a robust solution for surgical instrument detection in complex laparoscopic scenarios.