Gauze detection and segmentation in laparoscopic liver surgery: a multi-center study

Area of Science:

• Medical Imaging
• Artificial Intelligence in Surgery
• Surgical Safety

Background:

• Surgical gauze is easily overlooked in laparoscopic procedures due to its size and low visibility.
• Omission of surgical gauze poses a risk to patient safety and surgical efficiency.

Purpose of the Study:

• To develop a deep learning framework for accurate gauze detection in laparoscopic liver surgeries.
• To enhance surgical safety and efficiency by minimizing the risk of retained surgical items.

Main Methods:

• Trained deep learning models on 33 laparoscopic liver surgery videos from two hospitals.
• Employed frame-by-frame detection and segmentation of gauze objects.
• Introduced a quantitative method to categorize gauze detection difficulty (easy, moderate, difficult).

Main Results:

• YOLOv8n model achieved high accuracy in gauze detection, with recall and precision up to 0.9103.
• FCN-ResNet101 model excelled in gauze segmentation, reaching Dice scores of 0.9389.
• Performance was validated across internal and external datasets and varying difficulty levels.

Conclusions:

• The deep learning framework demonstrates robust capability in detecting and segmenting surgical gauze in complex laparoscopic liver surgery videos.
• This AI-driven approach shows significant potential to improve surgical gauze management and patient outcomes.