Every year, thousands of families are forced to grapple with a heart-wrenching realization-someone they love has dementia. In the United States alone, more than 5 million adults over the age of 65 live with dementia, and that number is projected to grow to 14 million by 2060. The most common form of dementia is Alzheimer's disease, which is caused by changes in the brain that are currently irreversible.
With doctors expected to see an unprecedented number of dementia patients in the coming decades, Boston University researchers are developing new technologies to assist them in diagnosing, predicting, and monitoring cognitive changes.
Because dementia can cause symptoms that vary in different people-including problems with memory, speech and communication, healthy reasoning, and other cognitive impairments-diagnosing it is a challenge for doctors. Two studies-one that uses artificial intelligence to predict dementia onset, and another that uses at-home technology to track a patient's progress-aim to help accurately diagnose a person's cognitive decline. Knowing whether a person has dementia or a different condition that is treatable can help doctors care for patients as quickly and efficiently as possible.
Predicting Alzheimer's with Artificial Intelligence
To diagnose dementia, physicians typically perform cognitive tests on attention, memory, problem-solving, and other abilities, along with a physical exam, blood tests, and brain scans. This battery of evaluations can help determine whether the memory issues a person is experiencing are caused by Alzheimer's or another form of dementia.
"Patients walk in and doctors have to try to understand where they fall on the dementia spectrum," says Vijaya B. Kolachalama, a BU School of Medicine assistant professor and expert on using computers to aid medical diagnoses. He and his team developed a deep learning algorithm that, for the first time, attempts to predict where a person falls on the dementia spectrum and identify if their memory loss is due to dementia or other reasons.
The computer model, which Kolachalama would eventually like to see used in hospitals and doctors' offices, makes predictions based on information about the patient, symptoms, cognitive test performance scores, and brain scans. The technology, supported by the Karen Toffler Charitable Trust, has been under development for years and has gone through a series of tests to validate the results. In 2020, Kolachalama's research team, including Rhoda Au, a MED professor of anatomy and neurobiology and director of neuropsychology at the Framingham Heart Study, found that the algorithm outperformed experts in making Alzhiemer's diagnosis. Their latest findings further this work, and are published in Nature Communications.
"We found that the model is clinically relevant," Kolachalama says, and is the closest any model has come to mimicking what happens in a real-life medical setting. "This paper underscores the clinical significance of what we did."
To test the accuracy of the model's predictions against the expertise of neurologists and specialists who interpret brain scans and cognitive test results, the research team did side-by-side comparisons of 100 diagnostic cases, putting the computer models in direct competition with clinicians. After going over each case, the team found their model identified a person with normal cognition, mild cognitive impairment, Alzheimer's, and non-Alzheimer's dementia with the same level of accuracy-or sometimes even better-as doctors.
"We want to make sure the tool can be embraced by physicians," Kolachalama says. Now, with promising results, he and the research team are planning to initiate another comparative study in a clinic. An expert would make their decision based on the standard tests, as would the model, and then they could compare results-a critically important step before potential clinical trials and a full real-world launch. In the future, Kolachalama hopes his technology can not only speed up the testing process, but also help fill gaps in care in parts of the country that have less access to specialists treating memory disorders.
"Boston, for example, has really good experts that can diagnose Alzheimer's, and since the model is doing that task, it could potentially minimize their clinical workload by serving as an assistive tool," he says. "There are places where people don't have access to MRI or other tools, and so our model can be more valuable in those settings."
Tracking Memory in the Long Term
While Kolachalama hopes to speed up the initial diagnostic process, a second project at BU aims to help doctors and patients better track cognitive decline over time. Andrey Vyshedskiy, a neuroscientist and lecturer at BU's Metropolitan College, developed the Boston Cognitive Assessment (BoCA), an innovative free online test people can take at home-and that gives results in minutes.
"There are many reasons a person can have memory or cognitive impairment," Vyshedskiy says, for example, a vitamin deficiency, thyroid hormone imbalance, or other medical problems. He and his team recently completed a control study of 400 patients, 50 with clinically diagnosed Alzheimer's or mild dementia, to validate the accuracy of the testing platform. The results are published in BMC Neurology.
BoCA, which can be taken on a computer or smartphone app, takes about 10 minutes to complete and can be used to monitor memory and cognition over the course of weeks to months. Questions test a person's recall and memory, motor skills, problem-solving, and other tests that are clinically shown to be helpful indicators of dementia or other memory issues. According to Vyshedskiy, it is the first free cognitive assessment that stores results for long-term monitoring, so a person can compare their results over a period of time; the BoCA test is different every time to eliminate any possible learning curve. The results of the repeated testing can then be examined by a doctor.
"People get really anxious about their memory and brain health, and so this is a way to check themselves and establish a baseline," he says. Although anyone can access the assessment, BoCA can also be administered by primary care physicians. But if a person is doing the assessment without supervision of their doctor, Vyshedskiy recommends they take the test on the same day at the same time each week for months at a time, and then follow up with a healthcare professional with any concerns.
Vyshedskiy says BoCA will be especially useful for doctors to monitor a patient's recovery after a surgery that involves anesthesia, or to assess a patient's mental state over the course of weeks or months while starting a new medicine or treatment. He would also like to see clinical researchers take advantage of the free testing platform, since BoCA can be used by researchers conducting clinical trials for therapies or treatments for people with dementia or Alzhiemer's to effectively monitor study participants over the course of a trial.
