Medical diagnosis identifies a likely condition from a patient’s symptoms, observations, and test results. For AI, the challenge is not merely to suggest a condition, but to support decisions that are useful in clinical care. section 8.0 section 6.2.1
An illustrative diagnostic sequence is:
- Address emergencies first, such as active blood loss. section 8.0
- Check an obvious diagnosis with a decisive test where appropriate, then provide treatment. section 8.0
- When the answer is unclear, test likely hypotheses using affordable, safe, noninvasive options before more burdensome ones. section 8.0
The COVID imaging test
A proposed shortcut was to train a machine-learning model on chest images labeled “infected” or “not infected,” then use uploaded scans for rapid diagnosis. Yet a review of 2,212 studies found no identified COVID imaging model suitable for clinical use, citing methodological flaws or underlying biases. A plausible demonstration is not the same as a reliable diagnostic tool. section 6.2.1
