Instrument tracking and analysis in minimal access surgery for surgical skill assessment.

For this project, we analyse cataract surgery videos. It is known that the motions of both camera and surgical instruments are indicative of surgical skill in simulated environments.Through the application of computer vision algorithms, we attempt to automatically measure these motions. Video data from cataract surgery videos have many sources of noise that complicate the observation of such motion. As no 'de facto' method exists for tracking surgical instruments we investigate the validity of applying cues based upon colour, shape and motion to identify surgical instruments. In addition, we develop a iris tracker based upon Histogram of Gradients object detection to measure the changes in camera state throughout a procedure. A methodology based upon invariant characteristics of surgical instrument motion is developed and applied to a large dataset of procedures. Metrics such as path length and number of motions for surgical instruments in cataract surgery are measured with this fully automatic methodology. Path length and number of movements are compared with surgeon's experience and skill level as measured with a manual surgical skill marking scheme. These metrics are shown to be proportional to a surgeon's experience and in agreement with manual measures of surgical skill