An African American man with Alzheimer's Disease holding his head
Credit: FG Trade/Getty Images

Boston University researchers have developed a machine learning based program that is able to predict progression of mild cognitive impairment to Alzheimer’s disease within six years based on analysis of a person’s speech.

Writing in the journal Alzheimer’s & Dementia, the researchers report that the model was able to predict whether or not a person would progress to Alzheimer’s disease with an accuracy of 79% and a sensitivity of 81%.

Early prediction of Alzheimer’s disease is important, as currently available treatments are more effective if started early. Knowing risk status also allows clinicians to make the right treatment plans and even assess if their patients are eligible for clinical trials, as there are many potential treatments in development.

Lead investigator Ioannis Paschalidis, a professor at Boston University and director of the Rafik B. Hariri Institute for Computing and Computational Science & Engineering, and colleagues developed the program using natural language processing with machine learning.

The team analyzed the speech of 166 people enrolled in the Framingham Heart Study with early signs of dementia. Of these, 90 had progressive and 76 stable mild cognitive impairment. The researchers had six-year follow up data of who progressed to Alzheimer’s and who did not.

“We wanted to predict what would happen in the next six years—and we found we can reasonably make that prediction with relatively good confidence and accuracy,” said Paschalidis in a press statement. “It shows the power of artificial intelligence.”

The researchers created the program using a combination of voice interview recordings and also added relevant factors such as age, sex, and education level. Some of the participant data helped create the program, which was then tested on a different collection of participant data.

“This study highlights the immense potential of integrating natural language processing techniques and speech data in predicting the future progression to Alzheimer’s disease. The method offers an opportunity to develop a cost-effective, widely accessible remote screening tool,” conclude the authors.

Paschalidis and colleagues are now planning to explore speech analysis for prediction of Alzheimer’s and dementia on a wider scale, for example using data from everyday conversations rather than formal interviews. They also want to add more data from daily life into the AI model to help improve accuracy.

Also of Interest