In the past decade, we have seen a myriad of fascinating and relevant machine learning and advanced analytics projects for healthcare coming both form the academia and the big tech industry surfacing on media channels and in specialised journals. Those projects bear the promise of a radical shift in medical practice, enabling for instance, less biased diagnostics encompassing an even larger amount of data sources, or truely personalised treatments. In addition to this, they’re often architected to function on commodity devices (e.g., phones, cloud) enabling wider access to the device and empowerment of patients. However it is too rarely that we see those promising endeavour making their way to the doctor’s office or in the patient’s hands creating a substantial missed opportunity. One can entertain the view that if healthcare could benefit directly from the pioneering machine learning tools available in labs, private or public, the healthcare system would leap ahead by at least 5 to 10 years. Yet this tsunami of new AI powered tools remains to be seen in the field. In this presentation, I will outline through projects recently carried out in the field of brain mapping for Neurosurgery and MRI mining for Mental Health, some of the key enablers for real world AI for patient care (as opposed as in the lab proof of concept), explain some of the pitfalls as well as limitations often encountered, and discuss perspectives and useful frameworks to consider patient safety and technological progress.