New AI Tells Doctors Which Brain Scan Matters Most for Alzheimer's

When doctors suspect Alzheimer's, they often use two types of brain scans: an MRI, which shows physical shrinkage or damage, and a PET scan, which shows how active brain cells are taking up sugar or whether they're clogged with amyloid plaques. Each tells a different part of the story. But until now, AI tools that read these scans simply gave one answer—"Alzheimer's" or "not Alzheimer's"—without explaining what the AI actually saw.

That's what this new study changes. Researchers built an AI system called MCNet that not only diagnoses the disease, but also calculates a "modality contribution score" telling doctors how much each scan drove the decision. For example, in one patient it might be 80% MRI, 20% PET. In another, the reverse. The score is different for every person, which means the AI can reveal whether a patient's condition is driven more by structural damage or by functional decline.

The findings suggest that as patients move from healthy to mild impairment to full Alzheimer's, the PET scan becomes more important. In their test with 327 people, the AI was accurate at staging the disease (AUC of 0.88). And that pattern held up when they checked the work against over 1,000 other patients from a separate dataset, which is reassuring for real-world use.

Why you should care: Alzheimer's affects millions, and early, accurate diagnosis is hard. This AI is not just another black box. It gives doctors a clearer, per-patient explanation of what's driving the disease, which could lead to better personalized care and smarter decisions about which treatments might work—or which clinical trials patients should join.