• ALFF
• anxiety disorders
• classification
• covariance matrices
• fALFF
• functional connectivity
• Hurst Exponent
• non-linear analysis
• ReHO
• resting state fMRI
• Riemannian geometry

Machine learning is increasingly being applied to fMRI. Those methods, fit for large dimensional datasets, seem particularly suitable for the challenge of decoding brain's intrinsic behaviour from resting state recordings (rs-fMRI).
The goal of this work is to find a classification algorithm able to distinguish patients with anxiety related disorders from healthy controls, by analyzing resting state fMRI data. This work explores the possibility of using Support Vector Machines as classifying algorithm and presents two different approaches to feature extraction. First of all, it tests the power of features obtained from a linear and non-linear analysis of the fMRI time series. Later, it investigates the employment of functional connectivity matrices to classify the patients. The positive semidefinite structure of covariance matrices needs a Riemannian geometry framework to be properly used with SVMs: this work tries to tackle the problem by building a reliable and handy algorithm for rs-fMRI datasets classification. To check the reliability of this novel technique, the Human Connectome Project open access data have been used in addition to the anxiety related ones.
Despite the amount of different settings used, only the most interesting results are showed in this work in order to explain and discuss the methodological challenges of whole brain rs-fMRI classification.