Video processing of ophthalmic image sequences

Abstract

As modern day technology continues to develop, there has been an increased emphasis on the efficient integration and use of Computer Assisted Surgery (CAS) systems in the medical field. Reading low level data from the Operation Room and extracting high level useable results by processing the data, for example, can benefit the surgeons. The aim of this project is to automatically process cataract surgery videos obtained from the OR, using image processing techniques, in order to evaluate surgeons. One measure of evaluation, proposed by this project is to segment the surgery video into parts based on the different procedures, thereby measuring the time and hence efficiency of the surgeon for the particular tasks. A Support Vector Machine (SVM) was trained based on the SIFT features of images from a given video using the Bag of Visual Words (BoVW) approach. The SVM was trained to detect the different tools present in the surgery, thereby identifying the surgical task. Upon the successful implementation of the framework, the SVM was trained and tested with help of a small sample set of two cataract surgery videos. Using different frames from the same videos for the training and testing, a recognition rate of roughly 98% was achieved. Overall, through the different tests, the feasibility of the framework was depicted.