Fig. 1.

Scheme of the new T2-w MS lesion detection pipeline. The preprocessing in both baseline and follow-up for every modality (T1-w, T2-w, PD-w, and FLAIR) consisted in ROBEX skull stripping, N4 bias field correction, and Nyúl histogram matching. For each modality, an affine transformation from baseline to follow-up was computed and the images were subtracted. Also, the images were non-rigidly registered to get a deformation field. Afterwards, the baseline and follow-up intensities, the subtraction values, and the DF features were used to train a logistic regression classifier. In the post-processing, the probabilistic maps were thresholded to obtain a binary segmentation where all lesions smaller than three voxels were removed.