Why?
Why does adding a tickbox to stuff like breast cancer treatments ('you consent to allow your scans and diagnosis to be anonymised and used as training data to better detect cancer in others'), collating say, a few million images etc per year, churning it through a machine learning algo, then checking whether unseen stuff results in more or less accurate diagnoses, on perhaps a smallish GPU cluster, cost 250 million and take two years to get started?
I could literally do this on my desktop right now, if I had the data.