Modeling and segmentation of surgical workflow from laparoscopic video

Area of Science:

• Medical image analysis
• Surgical workflow modeling
• Robotic surgery

Background:

• Analyzing surgical procedures using automated operating room (OR) signals is gaining traction for workflow analysis, skill assessment, and context-aware OR development.
• Extracting meaningful information from laparoscopic videos in minimally invasive surgery is challenging due to high dimensionality.

Purpose of the Study:

• To develop a method for dimensionality reduction of surgical video features using tool usage information.
• To extract semantic information from laparoscopic videos for surgical modeling.
• To evaluate statistical models for surgery phase segmentation based on reduced feature spaces.

Main Methods:

• Proposed a method combining Canonical Correlation Analysis (CCA) for dimensionality reduction with tool usage information.
• Projected high-dimensional image features into a low-dimensional space to extract semantic information.
• Compared two statistical models for surgery modeling in the reduced feature space.
• Evaluated surgery phase segmentation using Dynamic Time Warping (DTW) in conjunction with CCA.

Main Results:

• Canonical Correlation Analysis (CCA) successfully reduced feature space dimensionality while preserving semantic information.
• The combination of Dynamic Time Warping (DTW) and CCA demonstrated superior performance in segmenting surgical phases.
• The proposed method enabled surgery phase segmentation using only video data.

Conclusions:

• The integration of tool usage information with Canonical Correlation Analysis (CCA) is effective for surgical video analysis.
• Dynamic Time Warping (DTW) combined with CCA provides a robust approach for surgical phase segmentation.
• This method holds promise for improving surgical workflow understanding and skill evaluation in minimally invasive surgery.