Korean research utilises LLM to predict dementia risk

A research team at the Korean government-backed Electronics and Telecommunications Research Institute in Daejeon, a city south of Seoul, has developed a predictive AI technology for screening mild cognitive impairment.

HOW IT WORKS

Developed by combining existing voice and text analysis models with a large language model, the integrated AI analyses recorded conversations of senior people to evaluate their cognitive function and predict their risk of developing dementia. It improves on similar existing AI models, which have been inaccurate in assessing an elder's thinking capabilities due to imprecise pronunciation and dialects, ETRI claimed. A person's speech provides insights that can be used to detect and predict declines in a person's cognitive, language, and speaking functions.

The AI reportedly achieved 87.3% accuracy in recognising Alzheimer's disease, a common form of dementia, based on a test using an open-access dataset of spontaneous speech from DementiaBank. 

Following this test, the predictive AI was incorporated into a mobile application, which is now being prepared for a trial at senior welfare centres.

The research team is also working to expand the LLM used in the AI to include images and videos, besides text and speech inputs. 

WHY IT MATTERS

By 2025, South Korea is projected to become a super-aged society where people over the age of 65 account for a fifth of the entire population. That figure can also rise to about 40% by 2050.

About 10% of the country's senior population or around 840,000 people are living with dementia, based on the latest South Korea Dementia Status report in 2021. Their number is expected to swell to over 3 million by 2050. 

ETRI's development of a dementia screening tool contributes to various efforts to prepare for such a future. 

The mobile application enables monitoring at the community level and thus helps in the early detection of mild cognitive impairment among the elderly. It also supports health management to halt the progression of dementia.  

THE LARGER TREND

A growing body of research highlights the use of AI in predicting people's risk of developing dementia, particularly Alzheimer's disease. Fujifilm announced the development of AI Technology for AD Progression Prediction in 2022. It was reported that the technology can predict the likelihood of a person's mild cognitive impairment progressing to Alzheimer's in two years. 

Eisai and Oita University in Japan also built what could be the world's first AI model that analyses wearables data to predict Alzheimer's. 

Meanwhile, another Korean organisation, the Korea Institute of Machinery and Materials, created an intraocular eye lens that reacts to an Alzheimer's biomarker, potentially enabling early detection.