Texas A&M University researcher Dr. Tianbao Yang is developing an artificial intelligence model designed to do something current diagnostic tools struggle with: determine not just whether a patient has Alzheimer’s disease, but precisely what stage of progression they are in. The project, detailed in a Texas A&M Stories feature published in July, trains the model on multi-modal patient data — combining brain imaging, cognitive test scores, and biomarker readings — to build a more granular picture of disease progression than any single test currently provides.
Alzheimer’s disease is typically staged using a mix of memory tests, functional assessments, and, increasingly, expensive PET scans or spinal fluid biomarker tests that many patients never receive until the disease is already advanced. Yang’s team is betting that AI’s core strength — finding patterns across large, messy, multi-source datasets — can substitute for some of that invasive and costly testing, giving clinicians an earlier and cheaper read on where a patient sits on the disease trajectory.
Why Staging Matters as Much as Diagnosis
Getting an Alzheimer’s diagnosis is only half the clinical picture; knowing the stage determines everything from medication eligibility to care planning. Newer disease-modifying drugs like lecanemab and donanemab are approved specifically for early-stage patients with mild cognitive impairment or mild dementia, meaning a delayed or imprecise staging can functionally exclude a patient from treatment options that only work before the disease progresses further. Yang’s model aims to close that window by identifying transitions between stages earlier and more reliably than clinical judgment alone.
The Access Problem Driving This Research
PET imaging and cerebrospinal fluid biomarker testing, the current gold-standard tools for confirming Alzheimer’s pathology, remain expensive, invasive, and unevenly available, concentrated mainly at academic medical centers and specialty memory clinics rather than the primary care and community neurology settings where most older adults are actually first evaluated for memory concerns. That access gap means many patients, particularly those in rural areas or without nearby academic medical centers, are staged using clinical judgment and cognitive testing alone, methods that are far less precise than imaging or biomarker confirmation. An AI model that can approximate biomarker-level staging precision from more widely available data, cognitive scores, basic imaging, and clinical history, could meaningfully narrow that access gap rather than simply adding a new tool for institutions that already have advanced diagnostic capacity.
A Parallel Breakthrough From USC
Texas A&M’s work lands alongside a separate advance from USC’s Leonard Davis School of Gerontology, published in the Proceedings of the National Academy of Sciences. Led by Associate Professor Andrei Irimia, that team trained a deep-learning model on MRI scans from nearly 15,000 cognitively healthy people to establish a baseline for how individual brain regions age, then tested it on more than 1,900 additional participants, including those with mild cognitive impairment and Alzheimer’s disease. Instead of reducing an entire brain scan to one aging score, the USC model maps aging voxel by voxel, revealing that the hippocampus, amygdala, and other deep cognitive regions age unusually fast in patients on the path to dementia — and that older local brain age tracked directly with worse cognitive performance.
Two Different Approaches, One Shared Goal
Where Yang’s model leans on multi-modal clinical data to classify disease stage, the USC model works purely from imaging to map regional aging patterns — but both are chasing the same unmet need: catching neurodegeneration earlier, when intervention has the best chance of slowing decline. Researchers on both teams describe this as a shift away from single-number diagnostic thresholds toward richer, more personalized disease models.
The Caution Behind the Optimism
Neurologists reviewing this generation of AI staging tools note that classification accuracy in a research cohort is not the same as clinical reliability across the diversity of real-world patients, many of whom have overlapping conditions like vascular dementia or depression that confound memory-based diagnosis. There is also a validation gap: long-term studies are still needed to confirm whether AI-predicted stage transitions actually correspond to how a given patient’s disease behaves over the following years, not just at a single snapshot in time.
What’s Next for AI Alzheimer’s Staging
Both research groups say broader, more diverse clinical validation is the immediate next step, along with efforts to make the underlying data — MRI scans, cognitive assessments, and biomarkers — more interoperable across health systems so models trained at one institution generalize to patients elsewhere. If these staging tools prove reliable, they could reshape how quickly patients are routed to newer anti-amyloid therapies, potentially extending the treatable window for a disease where timing has historically determined whether a drug can help at all.
