Network analysis of brain functions of patients with severe chronic disorders of consciousness using functional magnetic resonance imaging

Abstract

The diagnosis and prognosis of patients with severe chronic disorders of consciousness (scDOC) are still challenging and a lot of misdiagnosis is evident. We aim to show new techniques to get a better insight into these disorders and introduce methods that can help in the diagnosis, all focusing on functional magnetic resonance imaging (fMRI) as modality. After a general introduction, an overview of fMRI is given. The physical and physiological basis is described. Furthermore, the most important experimental designs as well as analysis techniques are explained. The next part focuses on scDOC patients and presents definitions of the different diagnoses and challenges that are faced nowadays. After these introductory sections, a chapter dealing with fMRI under anesthesia follows. The patient group of scDOC patients is especially difficult to investigate because they often cannot lie still for the duration of an fMRI scan. We show for the first time, as far as known to us, results of fMRI under anesthesia of a patient in minimally consciousness state. Such results are possible because we choose a specific task, namely stimulation with a brush. Moreover, in a second example we show that it is possible to detect brain activity in the motor cortex of a patient in the final stage of Creutzfeldt-Jakob disease when anesthetized. This is a new result because, up to now, it was assumed that patients in the late stages of Creutzfeldt-Jakob disease are in the apallic syndrome and thus do not activate the cerebral cortex no more. The next part deals with the analysis of resting state fMRI. Resting state fMRI describes an fMRI experiment when the subject does not perform any task during scanning. This is of special interest for the considered patient group because no active participation of the subject is needed for such an examination. We present different analysis methods dealing with regions of interest (ROI). The first is an ROI-to-ROI analysis using a special connectivity analysis software. After that we turn to a method using graph theory. We construct networks out of the correlation matrices of the ROIs and then analyze the modularity with the multislice modularity approach. We choose this method because standard methods are not able to detect differences in the modularity of the different patient groups with scDOC. We are the first to use this new technique on this patient group and are able to detect differences between the subgroups. Furthermore, we introduce a new approach for a classifier which is based on modularity detection with the multislice method. The last part of this chapter combines a genetic algorithm and a support vector classifier to find the ROIs which differ most when considering the diverse groups of scDOC patients as well as healthy controls. The approach to combine a genetic algorithm and a support vector classifier is well-established but it has never been used in this special way and with this patient group. The last section gives a conclusion and outlook of what kind of work still has to be done in this field and shows how our new results can contribute to the understanding of this disease, e.g., by showing which ROIs are the most important indicators for the different types of disorders.