Activity Detection and Skill Assessment for Dexterous Motions in Robotic and Minimally-Invasive Surgery

Activity Detection;Surgical;Skill Assessment;Motion;Trajectory;Encoding;Residents;Learning;Frenet Frames;String Matching;Computer Science

Ahmidi, NargesHager, Gregory D. ;

Johns Hopkins University

关键词: Activity Detection;    Surgical;    Skill Assessment;    Motion;    Trajectory;    Encoding;    Residents;    Learning;    Frenet Frames;    String Matching;    Computer Science;   

Others  :  https://jscholarship.library.jhu.edu/bitstream/handle/1774.2/44608/AHMIDI-DISSERTATION-2015.pdf?sequence=1&isAllowed=y

瑞士|英语

来源: JOHNS HOPKINS DSpace Repository

【 摘 要 】

Many years of clinical research have provided evidence that poor quality of care (associated with high mortality and disability) and physicians;; low skill levels are correlated. Recently, the Accreditation Council for Graduate Medical Education (ACGME) mandated medical residents to be evaluated continuously (not merely periodically) using competency-based measures. This prompts the development of systems and platforms that can help with automated skill analysis of surgical activities. The goal of this dissertation is to develop systems that can detect and classify a surgical task and automatically assess surgical skill using the signals collected from surgical motions. Such systems could measure surgical performance objectively, help trainees to manage their time efficiently, provide feedback to improve their skill, and make life-long learning easier to implement. We study the characteristics of surgeons;; tool motions from a wide variety of surgical activities. We show that the complexity of these surgical signals -- such as environmental parameters, operator handedness, and preferences of different surgeons -- challenges most state-of-the-art techniques for time-series analysis and classification. However, we hypothesize that humans performing dexterous tasks follow a sequence of identifiable recurring motions (motifs) with some variability. We describe a compact, robust signal representation that reduces signal complexity and variability, creating a more stable discrete rendering of movement. We then introduce a string similarity function to compare the resultant motion trajectories based on repeated string motifs. Skill level (or activity type) is then predicted for a given motion trajectory by comparing it against trained motif dictionaries of different classes. We have validated our proposed technique, and report results substantiating achievement of the goals of this proposal (activity detection and technical skill assessment), for different manipulation or surgical tasks, for structured and unstructured tool motions, and on different datasets (robotic, endoscopic, open surgeries, and human activities).