AI System Predicts When Alzheimer’s Patients Will Experience Cognitive Decline

A new artificial intelligence (AI) model created by MIT Media Lab researchers can help in predicting which patients at risk for Alzheimer’s disease will experience significant cognitive decline. By predicting these patients’ cognition test scores up to two years in the future, this machine learning platform could be used to improve the Alzheimer’s drug research and development process. These researchers will be presenting their findings next week at the Machine Learning for Health Care conference.

Shortcomings in Current Alzheimer’s Research

The selection of potential drugs and participants for clinical research regarding Alzheimer’s treatment has been an expensive, unsuccessful process thus far. Major pharmaceutical companies have spent hundreds of billions in researching the disease over the past 20 years, but these efforts have experienced many failures.

A 2018 report from the Pharmaceutical Research and Manufacturers of America states that there were 146 failed attempts of Alzheimer’s drug development between 1998 and 2017. Only four new medicines were approved in that time, each functioning to alleviate symptoms of the disease. There is still a strong effort to develop these drugs, with over 90 treatments currently under development.

Research indicates that the best route of treatment may be recruiting candidates in the disease’s early stages when symptoms have not yet begun. Though many solutions have been proposed to detect Alzheimer’s in its earliest stages, such as the use of virtual reality, it is difficult to diagnose the condition before significant cognitive decline takes place. By using AI, however, these MIT researchers have created a tool that can help physicians find this specific patient population.

How the AI System was Developed

This research team first trained the system with an Alzheimer’s patient dataset, consisting of significant cognitive test scores and other biometric data. Information from healthy patients was included as well, with both sets of data being taken between biannual physician checkups. The machine learning model learns patterns from this data that help in predicting how patients will fair on cognitive tests between visits. For the new participants, a second model is used that continuously updates score predictions from newly recorded data in their more recent doctor’s visits.

This data was accessed through the Alzheimer’s Disease Neuroimaging Initiative (ADNI), the world’s largest Alzheimer’s clinical trial dataset. This source includes data from roughly 1,700 participants with and without the disease, taken at physician checkups over 10 years. Data in the set includes cognitive scores, MRI scans, cerebrospinal fluid measurements, and genetic and demographic information. The researchers personalized the population model for each individual participant and created a novel “metalearning” scheme to improve accuracy.

Accurately Predicting Patients at Risk

The researchers found that this model can make accurate predictions for these patients’ scores six, 12, 18, and 24 months in advance. By using this model, physicians could identify the patients at immediate risk for cognitive decline, who clinical trials are aiming to recruit. Treating these patients in the early stages of the disease will help the medical community find which medications work best.

“Accurate prediction of cognitive decline from six to 24 months is critical to designing clinical trials,” says Media Lab researcher Oggi Rudovic. “Being able to accurately predict future cognitive changes can reduce the number of visits the participant has to make, which can be expensive and time-consuming. Apart from helping develop a useful drug, the goal is to help reduce the costs of clinical trials to make them more affordable and done on larger scales.”

The researchers plan to form partnerships with pharmaceutical companies to put this AI model in place for Alzheimer’s clinical trials. Rudovic feels this model can be used to predict a variety of information outside of cognitive scores for Alzheimer’s and other diseases.