Magnetic Resonance Imaging
The group does research in machine learning applications in magnetic resonance imaging (MRI) of the human brain. One application is the spectroscopic MRI, which aims to detect metabolite levels in the brain using spectral methods, which is useful for the characterization of certain mental diseases.
In this application it is necessary to reduce the influence of the lipid spectrum around the brain to prevent this blurring the metabolite spectrum. This is achieved by positioning field bands on lipids to increase its resonance frequency and thus move its spectrum away from the metabolites. The placement process is manual, and only good results in two dimensions are achieved. besides, it is very tedious and time consuming in three dimensions, which should be placed up to 16 bands around the brain.
G2PI, in collaboration with the MIND Imaging Center (USA) and other centers, developed a technique to fully automatically and quick positioning between 6 and 16 bands of magnetic field, achieving high efficiency through an objective quality measure. We developed a graphical application in Matlab (left image) and carried out experiments in vivo MRIs in three dimensions, something that had not been achieved so far. The results and proposed algorithms can be seen in the papers of Martinez et al, 2010 and Yung et al., 2011, both in Magnetic Resonance in Medicine.
Subsequently, the consortium published a development capable of performing these tasks in real time (Zheng, 2012). It also developed a graphical application to carry out the training and testing process.
The group has experience in the use of learning techniques for classification and identification of mental disorders such as schizophrenia and obsessive compulsive disorder (OCD). In both applications techniques are used for determining which areas of the brain contain information about the pathologies, using fMRI multimodal images (schizophrenia) (see Castro et al. 2011, Neuroimage) or structural MRI images (sMRI) (Parrado et al., in preparation). Schizophrenia is characterized by a set of structural and functional areas from standard maps (MNI AAL). In the case of OCD, vocel selection techniques are used to detect approximately 40 brain areas involved in the disorder, which characterize patients and are consistent with previous clinical knowledge about the pathology.
En el ámbito del cartografiado cerebral (conocido como Brain Mapping), se está trabajando en alternativas a los métodos estándard basados en el SPM de Friston. Estos métodos alternativos están basados en el uso de técnicas de información mutua (véase Gómez-Verdejo et al, 2012, Medical Image Analysis), independencia estadística, procesos gaussianos y núcleos de Mercer en general.
In the field of brain mapping, we are working on alternatives to standard methods based on Friston's SPM. These alternative methods are based on techniques using mutual information (see Gómez-Verdejo et al, 2012, Medical Image Analysis) statistical independence, Gaussian processes and general Mercer's kernels.