Machine Learning Based Image Registration
Oral Defence Date:
Wednesday, May 12, 2010 - 15:30
Profs. Okada, Petkovic & Singh
This thesis proposed a novel machine learning based image feature selection solution for medical image registration. Image registration is a process of transforming different sets of data obtained from different measurements into one coordinate system for comparison or integration. In proposed solution, machine learning technique was used to train feature detectors in a cross validation process within our 3D CT scan training data set. The feature detectors with high cross validation score were then selected and used upon our 3D CT scan testing data set to do image registration. Furthermore, quantitative evaluation of proposed method is performed in the context of feature selection and medical image registration. Results of quantitative experiments suggest that performance improvement in both accuracy and stability has been achieved by introducing proposed machine learning based feature selection method.
Feature selection method, 3D image registration, 3D medical image processing, machine learning