Automated surgical step recognition in normalized cataract surgery videos

Area of Science:

• Ophthalmology
• Computer Vision
• Medical Informatics

Background:

• Vast amounts of surgical data are generated during video-monitored surgeries.
• Content-based video retrieval systems aim to leverage this data for computer-aided surgery applications.
• Accurate recognition of surgical steps is crucial for effective video retrieval and analysis.

Purpose of the Study:

• To develop a real-time system for recognizing cataract surgery steps.
• To improve the performance of content-based video retrieval systems for surgical videos.
• To refine motion feature extraction by addressing limitations caused by eye motion and zoom variations.

Main Methods:

• Utilized motion features for video characterization in cataract surgery videos.
• Implemented a novel pupil center and scale tracking method for pre-processing motion features.
• Evaluated the effectiveness of pre-processing on two different motion features.
• Applied a similarity measure adapted from video surveillance systems to a surgical dataset.

Main Results:

• The novel pupil tracking pre-processing significantly improved motion feature extraction.
• Retrieval performance was significantly enhanced for both evaluated motion features.
• The adapted similarity measure demonstrated good results on the surgical dataset.
• The system achieved significant improvements in retrieving videos of the same surgery step.

Conclusions:

• The proposed pupil tracking pre-processing method effectively refines motion features for surgical video analysis.
• This approach enhances the performance of content-based video retrieval systems in the context of cataract surgery.
• The study highlights the potential of advanced computer vision techniques for computer-aided surgery.