Robotic Surgical Skill Assessment Based on Pattern Classification Tools
Naikavde, Yuvraj Ratan
MetadataShow full item record
An improved understanding and quantitative assessment of human performance of dexterous manipulation tasks, especially in conjunction with tool, is a challenging area of research. This process gains importance as robots begin to work hand–in–hand or replace humans in the performance of such skilled manipulation tasks. In area of surgery robotic minimally invasive surgery (RMIS), has proved to be very beneficial by enhancing the stability, precision, dexterity and manipulation accuracy of surgeons. The da Vinci surgical robot has proved to be great success in the field of laparoscopic surgery. New experiments and research in the field of RMIS have shown that it is also a powerful tool for training, learning, and skill assessments. In order to understand the skill there is a need of sturdy and clear skill assessment technique. Skill assessment in various fields is based on how are the small sub tasks performed, which are judged and graded by experts in that field. This manual process is very time consuming, costly and esoteric. There is a need of automation in this process. The technique of distinguishing of an expert from novice is related to the way a big task is divided into small sub tasks. If this process is automated it can be helpful for training and also building blocks for training and automatic assessment. Past research in the field of Skill assessment in RMIS has centered with simple statistics on time taken to complete overall task, task segmentation, HMM model making for each skill level, however there is still no study done in the building state machines (reflecting features such penalty for breaking a standard sequence) on skill levels, based on the subtask performed which could give feedback relating to question like how novice can improve to become expert. This work addresses two main issues in the skill assessment. I) To classify different task performed with help of raw data and form a feature vector. II) To build a pattern recognition method which gives us the skill level based on the different tasks used to perform the whole activity. This thesis addresses the first task by collection of kinematic tool data collected from two expert surgeons, two intermediate surgeons and two novice surgeons. This data is divided into various sub tasks with help of a fixed threshold on velocity, position; region based tagging and video analysis. The second task of correlating sub tasks to the skill is carried forward with a sub task motion matrix in which the task specific data is encapsulated in a matrix (collection of feature vectors) which are used with the help of pattern classification tools (Naive Bayes, KNN classifier, Neural Network, SVM) to learn the skill level and then predict the skill for the test data. With basic 9 features could classify novices 100%, intermediate 67% and experts 70% for the pick and place task exercise and 90% experts, 87% intermediate and 100% for novice in peg board data. With high classification accuracy, features can be mapped back to skill, which can give feedback regarding which features are poor and how much distant are they from expert's performance. This work intends to give a platform where skill assessment can be given as a feedback for training novice surgeons or to learn the skill for training a sophisticated surgical robot. This method will give automatic feedback with less time and effort, reduce training cost and will improve success rate of surgery. It will increase learning rate and help trainees to learn which subtask they are week and need improvement. Application of this method can be applied in various other fields where skill learning can boost up the performance in both, humans and robots. Keywords: Surgical skill assessment, Motion analysis, Sub task, Pattern classification.