The signatures or image-based biomarkers are generated from a training set or library. To apply for an actual patient, the MRI may vary from the training set in terms of patient size, MRI scanner, MRI scanning pulse sequences etc.. To account for that, the signatures must be transformed. To illustrate

The transformation uses the statistics of a common object such as the normal prostate organ in both the training and the actual patient. Such a transform generates more accurate Gleason scoring as evidenced in the Preliminary Study.