Texas A&M Trains AI to Identify Alzheimer’s Stage Using Multimodal Patient Data

Researchers are developing a system that uses patient data to help physicians identify the disease sooner.

July 15, 2026 — Artificial intelligence excels at spotting patterns in large data sets, and Texas A&M University researchers are aiming to leverage that strength to detect Alzheimer’s disease sooner.

There is currently no single test that can determine if a person has Alzheimer’s. According to the Alzheimer’s Association, physicians typically use a combination of exams, including brain imaging, cognitive testing and blood tests, to make an accurate diagnosis. Dr. Tianbao Yang, professor and holder of Stephen Horn ’79 Engineering Excellence Chair position in the Computer Science and Engineering department, is training an AI model to analyze a patient’s multi-modal data and determine what stage of Alzheimer’s disease they may be experiencing, alongside Dr. Israel Liberzon, a professor in the Department of Psychiatry and Behavioral Sciences at Texas A&M Health.

The model is also being trained to explain the reasoning behind its decision. By identifying Alzheimer’s disease or risk early, physicians can develop targeted treatment plans to slow its progression and increase a patient’s quality of life.

“With Alzheimer’s disease, symptoms are often observed very late in the process. A person could have Alzheimer’s for 10 years before observing their first symptom,” Yang said. “For early stages, we can use certain therapies and medicine to help mitigate and slow down this process. In later stages, that’s a different treatment plan, which is why it’s important to know what stage of Alzheimer’s a person is experiencing.”

When determining Alzheimer’s risk or diagnosis, specialists often use tests like the Mini-Mental State Examination (MMSE) or the Montreal Cognitive Assessment (MoCA). They may also consider genetic information, age and education level. The AI model is trained on a public data set of 10,000 patients provided by Alzheimer’s Disease Neuroimaging Initiative.

Yang said it will be important for the AI model to not only diagnose Alzheimer’s disease, but also to explain how it reached its conclusion, allowing physicians to ensure accuracy. They developed a framework that automatically curates reasoning data based on symbolic constraints, enabling the training of highly faithful reasoning models that are better than traditional deep learning models.

“For neurologists, it may take many years of training to have the expertise to diagnose Alzheimer’s disease, especially in the early stages when it looks similar to other mild cognitive impairments,” he said. “For example, the model could make its decision because of MRI images, or a certain region of the brain where atrophy levels are high, but those rationales are key to making sure everything is accurate.”

Yang and his team are also exploring how the model could improve understanding of Alzheimer’s.

“Physicians already know many important features to look for, but they are also interested in whether AI can find new patterns in people with Alzheimer’s,” Yang said. “By looking at how different features interact across many patients, AI may reveal signals that could be useful for earlier and more precise diagnosis.”

In the future, Yang hopes to use the AI model to assess risk for other conditions, including stroke and heart disease.

“Physicians have certain templates and patterns they’ve been trained to look at in certain patterns of human data,” he said. “But in reality, the combination of different results and phenotypes complicates things and can be difficult for a human to capture. AI is good at finding these interactions between different features. We want this AI to speed up the diagnosis process with precision so people can get the care they need sooner rather than later.”

The project is funded by the Texas A&M Health Dementia and Alzheimer’s Research Initiative (DARI).

---

Source: Emma Lawson, Texas A&M University

The post Texas A&M Trains AI to Identify Alzheimer’s Stage Using Multimodal Patient Data appeared first on AIwire.