Pattern recognition and machine learning for magnetic resonance images with kernel methods

摘要

The aim of this thesis is to apply a particular category of machine learning andudpattern recognition algorithms, namely the kernel methods, to both functional andudanatomical magnetic resonance images (MRI). This work specifically focused onudsupervised learning methods. Both methodological and practical aspects are describedudin this thesis.udKernel methods have the computational advantage for high dimensional data,udtherefore they are idea for imaging data. The procedures can be broadly divided intoudtwo components: the construction of the kernels and the actual kernel algorithmsudthemselves. Pre-processed functional or anatomical images can be computed into audlinear kernel or a non-linear kernel. We introduce both kernel regression and kerneludclassification algorithms in two main categories: probabilistic methods andudnon-probabilistic methods. For practical applications, kernel classification methodsudwere applied to decode the cognitive or sensory states of the subject from the fMRIudsignal and were also applied to discriminate patients with neurological diseases fromudnormal people using anatomical MRI. Kernel regression methods were used to predictudthe regressors in the design of fMRI experiments, and clinical ratings from theudanatomical scans.