No single non-invasive test can diagnose Alzheimer’s disease. Providers rely on a combination of symptoms, health history, cognitive assessment, brain imaging, and psychological evaluation to piece together a picture from which they can make a reasonable diagnosis. A research team at the Stevens Institute of Technology has announced a potential breakthrough in the search for a more definitive diagnostic system: an AI algorithm that can identify Alzheimer’s disease with 95% accuracy.
Recent developments in computerized testing complement standard diagnostics because they’re standardized and repeatable. However, these options remain subjective and can only assess within a narrow category, such as attention or memory, so they’re essentially just another tool in the subjective diagnostic toolkit. The new algorithm uses granular data based on common language substitutions made by people with Alzheimer’s to analyze an individual patient’s language patterns.
Lead by K.P. Subbalakshmi, founding director of the Stevens Institute for Artificial Intelligence, the team developed a convolutional neural network — a type of AI that focuses on classification — equipped with attention mechanisms that allow it to learn from its own results. The researchers then converted existing texts created by Alzheimer’s patients and healthy individuals to numerical data. They used this data to train the network to recognize language patterns exhibited by people with the disease.
The AI software uses its algorithm to analyze patient speech, looking for markers that indicate Alzheimer’s disease. It learns by evaluating its own results and can in fact explain its own conclusions. That allows providers to verify the algorithm’s results before accepting a diagnosis.
Designed to incorporate ongoing discoveries in Alzheimer’s research, the team can easily input new data to generate even more accurate results. The next step for Subbalakshmi and her team is to expand the system to languages beyond English. The team also sees promise in exploring language analysis as a tool for insight and diagnosis of other neurological conditions, like stroke, traumatic brain injuries, and mental health conditions.