Computer Vision
Also called: image analysis, CV
The field concerned with extracting information from images, and the technical basis of nearly all authorized medical imaging AI.
Computer vision covers detection (is there a finding), localization (where), segmentation (which pixels belong to it), and classification (what kind). Medical imaging AI uses all four, frequently in combination.
Imaging dominates the FDA device list for structural reasons: it is already digital, already standardized through DICOM, and has measurable endpoints a regulatory submission can be built around.
Where This Gets Misread
Medical images are not photographs. Windowing, reconstruction kernels, slice thickness, and contrast timing all change the pixels substantially without changing the anatomy. A model trained on one institution's protocols can degrade on another's while the images look identical to a clinician.