A domestic research team has developed an artificial intelligence (AI) technology that analyzes the speech of the elderly to screen for mild cognitive impairment, a pre-dementia stage, and a high-risk group for dementia.

The Electronics and Telecommunications Research Institute (ETRI) announced on the 1st that it is conducting AI technology research to evaluate and predict degenerative brain function decline such as mild cognitive impairment and dementia by analyzing the voice utterances of the elderly input through conversations in daily life with support from the National Research Council of Science and Technology.

Speech production is a complex process in which cognitive functions such as memory, intention, and attention; language production functions such as phonemes, syntax, and meaning; and speech motor functions such as breathing, articulation, and vocalization work sequentially.

Therefore, through this speech analysis, it is possible to make early judgments and predictions on the decline in cognitive, language, and motor abilities in patients with mild cognitive impairment and dementia.

ETRI's Complex Intelligence Research Lab is expanding its research into the healthcare field, including digital therapeutics, based on AI technology accumulated in the field of speech processing and voice, text, and video multimodal technologies.

The research team combined existing voice and text analysis technologies with the world's first large-scale language model (LLM) to predict Alzheimer's dementia, and announced results from the ADReSSo Challenge dataset hosted by the University of Edinburgh in the UK and Carnegie Mellon University in the US.It achieved the highest performance of 87.3%, surpassing the previous record of 85.4%.

The research team's results were published in the ETRI Journal in February 2024.

Immediately after publication in the journal, it received a lot of attention, including inquiries from companies in the United States and Germany about the possibility of commercialization.

In a follow-up study, the research team applied the recently spotlighted visual language model (VLM) technology and renewed the best performance in the same ADReSSo challenge.

A paper has also been submitted to a top SCI journal.

Based on the research results, we completed the development of a tablet-based app that predicts high-risk groups for mild cognitive impairment through voice input centered on daily life conversation tasks.

There are difficulties in analysis due to imprecise pronunciation and dialect speech, which are common in the elderly, especially in those at high risk for mild cognitive impairment and dementia, but these have been overcome based on accumulated voice and multimodal AI technologies.

It was developed with a focus on improving user convenience and accuracy for the elderly, who are the actual users, and is planning to conduct verification at senior welfare centers in cooperation with the Korea Electrotechnology Research Institute.

ETRI's Complex Intelligence Lab's Senior Researcher Byung-ok Kang said, "Compared to the existing method of visiting a public health center in person to receive screening tests, the conversation-based screening method using smart devices has the advantage of enabling continuous/periodic monitoring."

The research team expects that this technology will help many elderly people at high risk of dementia to identify mild cognitive impairment early on and to delay the progression to dementia as much as possible through continuous management from the early stages, thereby greatly contributing to solving the most serious problem of dementia in our super-aging society.

This research result is evaluated as opening a new path for dementia prevention and early diagnosis through the fusion of AI and medical technology.

incenseIt is expected that through post-commercialization, it will reduce national and social costs for dementia treatment and have a big impact in the global digital therapeutics market.

This achievement was carried out as a project of the National Research Council of Science and Technology's creative convergence research project, 'Development of AI-based degenerative brain function decline evaluation technology through the establishment of big data on daily life speech of the elderly.'