The approach is good for getting quick images where pattern recognition is able to 'fill in the blanks'. But in medicine the whole point is that you are looking for something pathological. Especially with tumors there is no really good segmentation algorithm that catches all shape types.

Relying on post processing and pattern recognition will cause uniquely shaped tumors to be missed (and I'm saying this as someone who has done his PhD developing medical segmentation algorithms. I love what they can do - but relying on them fully in medical diagnostics is a recipe for disaster)

Humans still have one thing where they are way ahead of machines. That's the ability to go: "Whoa, what is that? Haven't seen one of those kinds before. Let's have a closer look."

Thanks for the comment. The technique involves a little more than post-processing. It is a fundamentally new way to acquire the scan as well. The technique was developed to get a better handle on disease states, like for example, observing white matter in demyelinating diseases.

Humans still have one thing where they are way ahead of machines. That's the ability to go: "Whoa, what is that? Haven't seen one of those kinds before. Let's have a closer look."

Exactly. That was how my first diagnosis (Diffuse Large B-cell Lymphoma) was made. As a matter of fact as soon as I felt it I remember using the "whoa" word.