AI diagnostics and what clinicians can verify in oncology

"The more AI changes what we do … the less able we are to check it.”

1

AI enhances diagnostic capabilities in oncology by identifying features in medical images that may be missed by human observers.
2

Dr. Kann outlined three tiers of AI applications, with tier one being the most verifiable and tier three presenting the greatest challenges.
3

Currently, there are no FDA-cleared tier three AI applications in radiology, though the Artera AI assay is a successful tier three application in oncology.
4

Verifiability of AI applications decreases with higher tiers, complicating the assessment of their effectiveness in clinical settings.
5

Clinicians must understand model outputs, including calibration and limitations, before integrating AI tools into patient management.

1