Neurodegenerative diseases often look alike in the clinic, especially early on, when Alzheimer's disease, Lewy body disease and other conditions produce overlapping patterns of cognitive decline. A team led by Jacob Vogel, assistant professor at Lund University within the MultiPark research area, with postdoctoral researcher Lijun An as first author, has developed an AI model that reads blood protein measurements to detect several of these conditions at once.
Working with the Swedish BioFINDER study and the Global Neurodegeneration Proteomics Consortium, whose database is the largest of its kind, the researchers trained the model on protein data from more than 17,000 patients and controls. Using a joint-learning approach, it identified a protein signature common to brain-degenerating diseases and distinguished conditions including Alzheimer's, Parkinson's, ALS, frontotemporal dementia and prior stroke. Published in Nature Medicine, the model outperformed previous approaches, was validated across independent datasets, and predicted cognitive decline better than clinical diagnosis did.
Perhaps the most provocative result was biological rather than technical: many people carrying an Alzheimer's diagnosis showed protein patterns closer to other brain disorders, suggesting mixed pathology, varied disease routes, or misdiagnosis. The authors are careful to note that blood proteins alone will not suffice for multi-disease diagnosis and that the method must be refined and combined with other clinical tools. Even so, a single routine blood test that screens across the dementias would transform how memory clinics triage patients.