Researchers have developed an artificial intelligence system that can detect multiple neurodegenerative diseases from a single blood sample, addressing one of medicine's most frustrating diagnostic challenges. A new study from Lund University in Sweden, published in Nature Medicine, describes an AI model called ProtAIDe-Dx that analyzes blood protein patterns to identify several brain conditions at once, including Alzheimer's disease, Parkinson's disease, and stroke .

Why Is Diagnosing Brain Disease So Difficult?

Memory loss and cognitive decline can point in multiple directions, making diagnosis messy and slow. A patient may seem to fit Alzheimer's disease, only to have signs that also resemble Parkinson's disease or a past stroke. The problem is especially acute because these conditions often overlap in the same brain. Misdiagnosis rates run around 25 to 30 percent even in specialized dementia clinics, and can exceed 50 percent in primary care settings . Among people 80 or older, the challenge intensifies: approximately 70 percent may carry multiple neurodegenerative pathologies at the same time .

This diagnostic confusion matters because it delays treatment, complicates care planning, and can lead patients down the wrong therapeutic path. A single blood test paired with artificial intelligence could help sort out that confusion and identify which diseases are actually present.

How Does the AI Blood Test Work?

Instead of looking for one disease marker at a time, the ProtAIDe-Dx system was trained on data from 7,595 different proteins measured in blood samples . The research team built their model using a "joint learning" approach, which allowed the system to learn shared patterns across several disorders while still producing separate probabilities for each one. This means the AI can flag more than one possible condition in the same person, rather than forcing a single diagnosis.

The study included data from 17,187 participants gathered across 19 research sites through the Global Neurodegenerative Proteomics Consortium, or GNPC . Importantly, the researchers used only proteomic information, meaning they did not feed the model demographic data, cognitive test scores, clinical diagnoses, or information about which research site the sample came from. This approach helps ensure the AI is detecting disease biology rather than relying on demographic shortcuts.

What Were the Accuracy Results?

ProtAIDe-Dx performed impressively across multiple conditions. The model achieved 95 percent balanced classification accuracy for amyotrophic lateral sclerosis (ALS), 92 percent for Parkinson's disease, 81 percent for Alzheimer's disease, 72 percent for frontotemporal dementia, and 70 percent for stroke or transient ischemic attack (TIA) . The model also produced area under the curve values above 0.8 for every task except stroke or TIA prediction, indicating strong overall performance when compared to other machine learning approaches like Random Forest and XGBoost .

One striking finding emerged when researchers looked beyond simple yes-or-no diagnostic labels. The protein profile predicted cognitive decline better than the clinical diagnosis did, according to the study's first author, Lijun An . This suggests that blood proteins may carry a broader signal of brain disease than researchers once thought.

What Do the Results Mean for Diagnosis?

The research revealed something unexpected: some patients labeled with Alzheimer's disease showed protein patterns that looked more like other brain disorders. This could mean several things. More than one disease process may be present in the same person, Alzheimer's may develop through different biological routes, or the original clinical diagnosis may have been wrong. The model also picked up signals in harder-to-classify groups, including people with mild cognitive impairment or subjective cognitive decline, suggesting the system may help identify underlying pathology earlier than standard diagnosis alone .

"Our hope is to be able to accurately diagnose several diseases at once with a single blood test in the future," said Jacob Vogel of Lund University, who led the study.

Jacob Vogel, Researcher at Lund University

When tested on an external dataset called BioFINDER-2, which included 1,786 participants with biomarker-supported diagnoses, the model's probabilities tracked with expected disease biology. Alzheimer's probabilities rose in some non-Alzheimer's cases that carried amyloid-beta and tau pathology, the hallmark proteins of Alzheimer's disease . Stroke probabilities rose with heavier white matter hyperintensity burden, a sign of brain damage from reduced blood flow .

Steps to Understanding Blood-Based Brain Disease Detection

• Protein Analysis: The AI system examines patterns across thousands of proteins in blood samples rather than looking for a single disease marker, allowing it to detect multiple conditions simultaneously.
• Clinical Validation: Researchers tested the model against established biomarkers like amyloid-beta, tau protein, and white matter changes to ensure the AI predictions matched actual disease biology.
• Comparison Testing: The system was evaluated against other machine learning models and on external datasets to confirm its accuracy and generalizability across different populations.

What Are the Current Limitations?

The researchers are careful not to oversell the system. Current blood protein measurements are not sufficient by themselves to diagnose multiple diseases, and the paper emphasizes this point repeatedly . Performance dropped when the team tested how well the model generalized from one research site to another, a sign that site effects and data differences remain a serious challenge . Some diseases were also harder to classify because of uneven sample distribution or because their clinical labels were not backed by biomarkers.

The authors also note that many brain-related proteins may never reach the blood in ways that are easy to measure, partly because of the blood-brain barrier, a protective layer that prevents most substances from entering the brain . Medication effects can also distort protein levels. And because many of the training diagnoses came from routine clinical work rather than autopsy or biomarker confirmation, some "false" predictions may not have been false at all .

However, ProtAIDe-Dx added significant value when combined with common clinical markers such as age, sex, cognitive test scores, brain imaging measurements, and plasma p-tau217 and plasma NEFL, which are blood-based markers of neurodegeneration . In the BioFINDER-2 sample, that combined model improved diagnosis, especially for non-Alzheimer's dementias . The system also separated patients by future cognitive decline more effectively than baseline diagnosis did .

What Happens Next?

This study does not deliver a simple blood test that can replace brain scans, spinal fluid tests, or expert clinical evaluation. What it does offer is a glimpse of a more practical future, one where a single blood draw could help doctors sort through overlapping brain diseases, identify patients who need follow-up testing, and catch hidden pathology earlier. That could matter for treatment decisions, drug trials, and care planning as more disease-modifying therapies move closer to routine use .

"The protein profile predicted cognitive decline better than the clinical diagnosis did," noted Lijun An, the study's first author.

Lijun An, Researcher at Lund University

The next step, according to Vogel, is to expand the proteomic markers, including through mass spectrometry, to search for disease-specific patterns that current tools may miss . As researchers continue refining this technology, blood-based AI diagnostics could transform how doctors approach brain disease, moving from guesswork and delayed diagnosis toward faster, more accurate identification of what's actually happening in the brain.