Medical AI is well-suited to complex input, personalised output and scaling through automation

AI is useful in situations where the input is highly complex, where it is desirable for the question to have a complex answer or where the hypothesis space cannot easily be constrained. An example of a highly complex input is imaging data, which may contain numerous features that can represent a large number of pathologies.

AI is good at providing outputs that are personalised to individuals and patient sub-groups, rather than population-level scoring systems that provide generalisable but imprecise predictions for the non-existent ‘average’ patient. (Example: Personalized predictions of patient outcomes during and after hospitalization using artificial intelligence)

AI can expand the scope of research by enabling automation of routine tasks. For example, a research group wanting to anotate a large dataset of computer tomography (CT) would require substantial time and cost with AI. But with AI a subset could be labelled, used to train a classifer and create labels for the remaining images. The same is also true for other data types, such as AI extraction of information from clinical letters using natural language processing.

References