Computational Intelligence Methods for Medical Image Understanding, Visualization, and Interaction

Abstract

Computational intelligence is well-suited to medical data analysis and visualization. This thesis focuses on exploring new directions for applying computational intelligence methods to medicine. First, novel ensemble methods were developed for analyzing 3D CT scans and multimodal data, thus establishing and improving capabilities for osteopenia diagnosis, regression-based modeling of bone mineral density, and identification of feature-disease relationships. Second, a clustering-based method was designed to automatically identify 3D structures in medical volumes and thus enable the automatic design of transfer functions for context-based volume visualization. Lastly, new one-class classifiers were proposed for biometric recognition and outlier rejection in a gesture-based surgical data access system. These advances in computational intelligence have improved and established novel capabilities in medical data analysis, visualization, and interaction. The contributions in this thesis can be applied to support clinicians in making medical decisions, or to simplify and automate medical tasks to reduce labor.