Alzheimer's Scans Have a Gray Zone. People in It Are Already Declining

Alzheimer's brain scans have long been read as a yes-or-no answer: either the scan lights up enough to be called "positive," or it doesn't. The middle of the curve, where amyloid is detectable but not overwhelming, gets folded into one side or the other. A new study of 512 patients at the Geneva Memory Center says that middle is its own category, with measurable cognitive decline and tau biomarkers behind it.

The paper, published in the Journal of Nuclear Medicine, is titled "Clinical and Biologic Characterization of a 3-Level Centiloid-Based Classification of Amyloid PET." An amyloid-PET scan uses a radioactive tracer that binds to the fibrillar amyloid plaques in the brain; the Centiloid scale, 0 to 100, lets clinics compare those scan intensities across tracers and machines. Zero means no amyloid signal; 100 means a level typical of moderate Alzheimer's dementia. The authors ran statistical models against fluid biomarkers and longitudinal cognitive scores to find natural cut points in the data: roughly 13 Centiloids for cognitive decline, 14 for early tau in the mesial temporal lobe, and 51 for advanced cortical tau spread.

Below 12 Centiloids, the cohort is amyloid-negative. Above 37, it is amyloid-positive, with tau spread through the cortex. In between, from 12 to 37 Centiloids, sits the new intermediate band, where early tau tangles start showing up in the mesial temporal lobe, a deep-memory structure that lights up early in Alzheimer's.

That intermediate band is the new finding. Of the 512 patients seen between 2016 and 2024, 63 fell into it, about 12% of the cohort. Their cerebrospinal fluid and plasma biomarkers for amyloid, tau, and neurodegeneration sat between the low and high groups, not normal, not fully Alzheimer's. On the Mini-Mental State Exam, a 30-point cognitive screener, both the intermediate and high groups declined measurably faster than the low group. The intermediate group's pace was described as moderate. It was decline, not noise. The cohort mixed cognitively unimpaired adults, patients with mild cognitive impairment, and patients with dementia, so the intermediate band was not a single diagnostic group; it was a state of the scan.

This is where the clinical stakes sit. The same amyloid-PET readout that puts a patient in the "positive" or "negative" bucket is the scan that, in many systems, decides who is eligible for the new class of amyloid-targeting therapies, lecanemab and donanemab, that clear amyloid and modestly slow cognitive decline. Someone whose scan lands at 15 Centiloids today gets the same binary report as someone at 60, and lands on the wrong side of the same eligibility line. The Geneva proposal is, in effect, asking clinics to re-read scans they are already taking.

The 13-, 14-, and 51-Centiloid cut points are derived from a single specialty memory clinic in Geneva, and the authors are explicit that they are a proposal, not a standard. The intermediate group is small (n=63), and its MMSE-decline estimate carries wider uncertainty than the larger low- and high-amyloid arms. External validation in other clinics, with other PET tracers and other populations, is the next step the field will look for. Treatment decision-making implications are forward-looking and contingent on that prospective work.

What changes immediately is the conversation, not the prescription. A patient who today gets a "negative" amyloid scan, but whose Centiloid number sits at, say, 20, can now ask whether the report should have read "intermediate." For a clinic that already uses quantitative Centiloid reads, the 3-tier scheme is a small software change. For one that still reports only positive or negative, it is a reason to start.