Multivariate and sparse signal processing techniques in multimodal neuroimage analysis for the identification of neurological alterations.

Abstract

The diagnosis of neurodegenerative diseases, particularly Alzheimer’s Dis- ease (AD) and Parkinsonian Syndrome (PS), has been significantly enhanced by the advent of Computer Aided Diagnosis (CAD) systems. These systems, lever- aging advanced computational methodologies, aim to automate the recogni- tion of neurodegenerative patterns characteristic of these diseases. This disser- tation presents a series of innovative methodologies that have been developed to address the challenges and nuances of medical image processing in the context of these diseases.

A cornerstone of this research is the application of Robust Principal Compo- nent Analysis (RPCA) to brain imaging. This technique facilitates the automatic computation of Regions of Interest (ROIs) in brain images, ranking them based on their diagnostic relevance. The sparse error matrix, derived from RPCA, has emerged as a pivotal tool in determining brain areas intrinsically linked to AD. Furthermore, the fusion of features from diverse image modalities, such as functional Positron Emission Tomography (PET) and structural Magnetic Res- onance Imaging (MRI) data, has been explored, yielding promising results in both exploratory analysis and classification tasks.

The challenge of feature extraction, especially in high-dimensionality data- sets, remains a significant hurdle in medical image processing. This research addresses this challenge through sparse representations of data, offering a solution to the curse of dimensionality. By combining specialized classifiers, this approach not only aids in classification but also provides insights into the progression of illnesses. Notably, while functional changes are evident in AD patients, structural alterations become more pronounced during the disease’s early stages.