Analyzing speech using the same AI algorithms as a trending new chatbot could act as a screening tool for Alzheimer’s disease, research shows.
The findings, reported in PLOS Digital Health, demonstrate how AI-driven speech assessment could assist with the early diagnosis of dementia and help tailor interventions towards individual needs.
The system deploys the third generation of a large, specific language model called Generative Pretrained Transformer (GPT-3), which is produced by San Francisco research and development company OpenAI.
The same company launched the ChatGPT last month, which has been making waves in the chatbot industry for combining deep learning with natural language processing to mimic realistic human conversations.
“We know from ongoing research that the cognitive effects of Alzheimer’s disease can manifest themselves in language production,” said co-author Hualou Liang, a professor in Drexel’s School of Biomedical Engineering, Science and Health Systems in Philadelphia, Pennsylvania.
“The most commonly used tests for early detection of Alzheimer’s look at acoustic features, such as pausing, articulation and vocal quality, in addition to tests of cognition. But we believe the improvement of natural language processing programs provide another path to support early identification of Alzheimer’s.”
Speech and language impairment is an important biomarker of neurodegenerative disorders such as Alzheimer’s and is present in 60–80 percent of people with dementia.
Currently, Alzheimer’s disease is primarily diagnosed through clinical assessment such as brain imaging and cognitive tests, which can be expensive and can involve lengthy evaluation.
The researchers examined the value of GPT-3 in this context, training it using speech recordings and “text embedding”, which focuses on the way words are used and language is constructed.
They report that text embedding generated by GPT-3 could distinguish people with Alzheimer’s disease from others and infer cognitive test scores solely from speech data.
GPT-3 performed better than two of the top natural language processing programs, with fewer missed cases than either.
Text embedding considerably outperformed traditional approaches using features from acoustic speech recordings such as pausing, voice strength and slurring, and was even competitive with fine-tuned models.
The GPT-3 proved to be almost 20 per cent more accurate in predicting the Alzheimer’s severity than the Mini-Mental State Exam (MMSE) test scores used to determine dementia severity.
“We show for the first time that GPT-3 can be utilized to predict dementia from spontaneous speech,” the researchers report.
They add: “Our AI model could be deployed as a web application or even a voice-powered app used at the doctor’s office to aid clinicians in [Alzheimer’s Disease] screening and early diagnosis.”