UTSW-led study finds an automated speech analysis algorithm is able to recognize cognitive impairment in a Spanish-speaking population

A novel speech analysis tool that uses artificial intelligence successfully detected mild cognitive impairment and dementia in a Spanish-speaking population, according to research led by UT Southwestern Medical Center. The study, published in Frontiers in Neurology, provides preliminary support for the algorithm as an early screening tool that may help identify patients at risk of developing dementia.

Dementia is an impaired ability to remember, think, or make decisions that mainly affects adults older than 65, but it is not a normal part of the aging process. Currently, an estimated 6.9 million people in the U.S. have Alzheimer’s disease, which is the most common cause of dementia. Early recognition of cognitive decline presents a clinical challenge but is a critical part of classifying patients with the highest risk of dementia.

“Analyzing a sample of speech obtained during some brief, routine neuropsychological tests shows promise in our ability to quickly screen for signs of cognitive impairment, particularly in population-based research studies. Machine learning-based tools such as this may play an increasingly important role in the future of cognitive screening for dementia,” said corresponding author C. Munro Cullum, Ph.D., Professor of Psychiatry, Neurological Surgery, and Neurology and Vice Chair and Chief of the Division of Psychology at UT Southwestern. Dr. Cullum is also an Investigator with the Peter O’Donnell Jr. Brain Institute.

Data for the study was collected from 195 Spanish speakers recruited as part of a multicenter clinical trial in Spain. All participants completed an initial evaluation and were categorized as either having normal cognition, mild cognitive impairment (MCI), or dementia. Data from 21 participants was excluded due to incomplete cognitive or demographic data, or poor audio transcription quality.

The final cohort of 174 participants had a mean age of 74; there were slightly more females (56%) than males. Participants were divided into a training group of 122 participants (70%) and a test group of 52 participants (30%).

Researchers used four language tasks to train independent machine learning (ML) models using data from the training group participants. Neuropsychological performance and audio recording variables were collected from each participant using the AcceXible platform – a proprietary web-based instrument developed for disease detection through speech analysis.

The final model of the speech analysis algorithm was then used for the test group and was able to differentiate cognitively normal participants from those with dementia or MCI with an overall accuracy of 88.4% and 87.5%, respectively. The final model outperformed one of the current standard-of-care screening measures known as the Mini-Mental State Examination (MMSE).

Findings from this study and similar work with English speakers by UTSW researcher Ihab Hajjar, M.D., Professor of Neurology and Internal Medicine and in the O’Donnell Brain Institute, suggest that these tools may improve quality of life for patients at risk for dementia through early detection – an issue that most significantly affects marginalized racial and ethnic groups who often experience delayed diagnosis. Further research is needed to validate the accuracy of the model before the technology can be deployed for clinical use.

“Eventually, such technology may help identify patients who are showing signs of cognitive decline that may be in need of clinical evaluation and consideration for treatment,” Dr. Cullum noted.