Reply to post:

Indeed and, while promising, training on 400 scans with a total of 728 lesions doesn't seem like nearly enough data to train on by several orders of magnitude, so very early days for this I would have thought.