Access Type

Open Access Embargo

Date of Award

January 2016

Degree Type


Degree Name



Industrial and Manufacturing Engineering

First Advisor

R. Darin Ellis


Surgery is continuously subject to technological innovations including the introduction of robotic surgical devices. The ultimate goal is to program the surgical robot to perform certain difficult or complex surgical tasks in an autonomous manner. The feasibility of current robotic surgery systems to record quantitative motion and video data motivates developing descriptive mathematical models to recognize, classify and analyze surgical tasks. Recent advances in machine learning research for uncovering concealed patterns in huge data sets, like kinematic and video data, offer a possibility to better understand surgical procedures from a system point of view. This dissertation focuses on bridging the gap between these two lines of the research by developing computational models for task analysis in robotic-assisted surgery.

The key step for advance study in robotic-assisted surgery and autonomous skill assessment is to develop techniques that are capable of recognizing fundamental surgical tasks intelligently. Surgical tasks and at a more granular level, surgical gestures, need to be quantified to make them amenable for further study. To answer to this query, we introduce a new framework, namely DTW-kNN, to recognize and classify three important surgical tasks including suturing, needle passing and knot tying based on kinematic data captured using da Vinci robotic surgery system. Our proposed method needs minimum preprocessing that results in simple, straightforward and accurate framework which can be applied for any autonomous control system. We also propose an unsupervised gesture segmentation and recognition (UGSR) method which has the ability to automatically segment and recognize temporal sequence of gestures in RMIS task. We also extent our model by applying soft boundary segmentation (Soft-UGSR) to address some of the challenges that exist in the surgical motion segmentation. The proposed algorithm can effectively model gradual transitions between surgical activities.

Additionally, surgical training is undergoing a paradigm shift with more emphasis on the development of technical skills earlier in training. Thus metrics for the skills, especially objective metrics, become crucial. One field of surgery where such techniques can be developed is robotic surgery, as here all movements are already digitalized and therefore easily susceptible to analysis. Robotic surgery requires surgeons to perform a much longer and difficult training process which create numerous new challenges for surgical training. Hence, a new method of surgical skill assessment is required to ensure that surgeons have adequate skill level to be allowed to operate freely on patients. Among many possible approaches, those that provide noninvasive monitoring of expert surgeon and have the ability to automatically evaluate surgeon's skill are of increased interest. Therefore, in this dissertation we develop a predictive framework for surgical skill assessment to automatically evaluate performance of surgeon in RMIS. Our classification framework is based on the Global Movement Features (GMFs) which extracted from kinematic movement data. The proposed method addresses some of the limitations in previous work and gives more insight about underlying patterns of surgical skill levels.