Abstract
      In the Brain Injury Service of the Hospital Valencia al Mar of Valencia there are patients who suffered a traumatic brain injury (TBI). The existing practice for diagnosing functional impairment in those patients was qualitatively evaluating FDG-PET imaging (Fluorodeoxyglucose Positron Emission Tomography) of the brain. In this study voxel-based analysis techniques is proposed for quantitative evaluation of FDG-PET images. For this purpose, an adaptation of the Voxel-Based Morphometry methodology is presented. The final aim of the quantitative evaluation of the images is that the method may serve as a support for the prediction of the degree of independence of the patients.
      Voxel-based analysis techniques were firstly applied to a single longitudinal case. In this case two FDG-PET images from a patient suffering from herpes encephalitis who had a relapse a year later were available, one for the first encephalitis and one for the relapse. A subtraction of PET volumetric images was performed, which was very useful to explain the different consequences of each episode. It also helped to confirm the hypothesis of a bi-phasic phenomenon, contrary to the progressive / degenerative hypothesis.
      In a subsequent study voxel-based techniques were applied at the group level to study the relationship between thalamic metabolism –as expressed in the FDG-PET imaging– and neurologic state for different groups of patients with TBI. This allowed us to acknowledge that metabolic differences between the various neurological states exist. Furthermore, the worse the neurological state of the patient was, the lower the thalamic metabolism remained.
      The recovery process of patients with TBI goes through various neurological conditions beginning with the coma state. Throughout this process the health personnel perform tests in order to see the neurological state of the patient. Depending on the state either a recovery plan is developed, or the current one is modified. This thesis presents a way to automatically classify the neurological state of patients with the only information of the patients’ FDG-PET images at rest, by using Support Vector Machines. The aim is that this method may be a support to health professionals in the classification of patients’ neurological condition for an optimal care. A study with three groups of patients and a control group is also presented. In this study the classification success rate in three of the four groups is 90 %.
      Finally, we describe the general conclusions of these studies, as well as future research guidelines.