A method for analysing a digital medical image of a subject is presented. The method comprises: - registering the medical image (10) with a reference template (12) using an affine 5 transformation (S1) - identifying one or more regions of interest (ROIs) in the medical image taking the registration of the medical image with the reference template into account (S4, S5) - encoding image data contained in the one or more ROIs into one or more 10 descriptors (32) indicative of the deviations of the image data from a generative model representing the statistical distribution of data of a training data set - based upon the one or more descriptors, generating a classification decision indicative of a condition of the subject
