Markov Modeling of Colonoscopy Gestures to Develop Skill Trainers*

摘要

Colonoscopy is a complex procedure which requires considerable skill by the clinician in guiding the scope safely and accurately through the colon, and in a manner tolerable to the patient. Excessive pressure can cause the colon walls to distend, leading to excruciating pain or perforation. Concerted efforts by the ASGE have led to stipulating guidelines for trainees to reach necessary expertise. In this paper, we have analyzed the motion of the colonoscope by collecting kinematics data using 4 electromagnetic position sensors. Further, 36 feature vectors have been defined to capture all possible gestures. These feature vectors are used to train Hidden Markov Models to identify critical gestures that differentiate expertise. Five expert attending clinicians and four fellows were recruited as part of this study. Experimental results show that roll of the scope shows maximum differentiation of expertise.