The present invention relates to a computer-implemented method for computing a neuromarker of Alzheimer's disease comprising the steps of obtaining at least one spectral feature from EEG signals of a subject； obtaining at least one Riemannian distance between a spatiofrequential covariance matrix computed from the EEG signals of said subject and at least one reference spatiofrequential covariance matrix； and combining said at least one spectral feature and said at least one Riemannian distance in a mathematical function. The present invention also relates to a method for self-paced modulation of EEG signals of a subject in order to alleviate symptoms of Alzheimer's disease using the predictive neuromarkers of Alzheimer's disease.
