I did my PhD research on a similar topic. I think the main utility wasn’t in diagnostics but rather often just telling the doctors to listen to the nurse (who flagged that something was wrong 4 hours before the model). I see a use for these tools to catch people who slip through the cracks, to act as a second opinion, and maybe to encourage some critical thinking, but I often get frustrated with this “AI is better than doctors” claim because it just glosses over how inconvenient and difficult medicine actually is in practice. (Disclaimer – I’m not a doctor. Or at least, the kind who helps people!)