Hi Krzysztof,

Samseg will "eat" anything you throw at it. That doesn't mean that the type of input data doesn't matter: the higher the contrast-to-noise ratio for the specific structure(s) you're interested in, the better the results will be. Note that you can often increase the contrast-to-noise ratio by giving several contrast combinations (i.e., multi-contrast input data) to samseg, although image quality issues (e.g., low resolution, imaging artifacts, ...) can have a negative overall impact. You may want to experiment with various combinations of your data to pick the best one suited for your problem...

Regarding the output, specifically for lesions the main output is the seg.mgz file, which contains a volumetric labeled image where every voxel has been assigned to a neuroanatomical structure (the level of lesions is 99). You can also save posterior probability maps for the various structures (incl. lesions), as explained on the wiki. This maybe be useful if you're interested in trading off false positive vs. false negative lesion detections differently from the default setting. FreeSurfer tools such as mri_binarize and mri_segstats can be used to obtain lesion volumes for different thresholds.