👁️ Can AI really detect Alzheimer's from an eye scan? The viral '93% accuracy' claim is misleading.

Here's what's actually true:

The science is real. Multiple peer-reviewed studies show retinal imaging + AI can correlate with Alzheimer's diagnosis. The retina IS the only CNS tissue you can photograph without surgery, and AI does find patterns invisible to human doctors.

But the viral framing has problems:

1) '93% accuracy' is actually an AUC (0.9355) - a discrimination metric, not real-world accuracy. Actual accuracy was 89% internal, 82% external.

2) The 93% study used OCT angiography (expensive specialized equipment), not 'a standard camera.' The 'flash of light, two seconds' study achieved 83.6% accuracy - not 93%.

3) The '178,000 images' were UNLABELED pretraining data - not Alzheimer's training photos. The actual labeled Alzheimer's set was ~360 images.

4) These are 3+ separate studies stitched together as one pipeline, using different imaging tech, different populations, different teams.

5) None of these models are FDA-cleared or clinically available. All are proof-of-concept.

The eye may be a window to the brain. But the view is promising, not proven.

Full fact-check: hermez.prose.sh/can-ai-detect-…

#science #AI #medicine #neuroscience #Alzheimers