Now Reading
How IBM Is Employing AI To Predict Alzheimer’s Disease

How IBM Is Employing AI To Predict Alzheimer’s Disease

  • This study is the first empirical evidence of 'tablet-based automatic assessments of patients using speech analysis'.
IBM Alzheimer's Research

Alzheimer’s is a progressive neurological disorder that develops with vague signals of mild memory loss that morphs into a slow but severe decline in cognitive functioning. More than 1 in 9 people of age 65 and older has Alzheimer’s disease. The US-based Alzheimer’s Association has predicted a 14 percent rise in Alzheimer’s prevalence by 2025. Without a definite cure for it yet, the best way to delay its onset is through early detection and intervention.

This year, a team of researchers from IBM and the University of Tsukuba developed an AI model to detect the onset of mild cognitive impairment (MCI) using surveyed persons’ responses to typical daily questions. MCI is a transitional stage between normal ageing and dementia. This study is also the first empirical evidence of ‘tablet-based automatic assessments of patients using speech analysis’.

Register for FREE Workshop on Data Engineering>>

Daily chats for assessment

The AI tool from IBM breaks away from the traditional process of analysing speech responses during cognitive tests to introduce a simpler model. The earlier studies focussed mainly on analysing speech responses during cognitive tasks, such as asking patients to describe a picture in detail.

IBM’s new AI model studies speech responses to daily life questions, some as basic as asking about their mood, plans for the day or what they had for dinner. This ensures daily monitoring with the AI embedded in smart speakers or smart home technology, warranting detection at the earliest. Thus, it is frequent with fewer assessment costs.

“We found that the detection accuracy of our tests based on answers to simple daily life questions data was comparable to the results of cognitive tests — detecting MCI signs with an accuracy of nearly 90 per cent,” IBM’s blog post stated. 

The technique

The study had a sample size of 76 senior Japanese persons, some with mild cognitive impairment (MCI) and cognitively normal people. The researchers then compared the speech feature’s results of both groups. 

It was a major challenge to capture subtle cognitive differences based on casual conversations with a low cognitive load. To overcome this, the researchers combined the responses to multiple questions and language functions associated with MCI and dementia. IBM’s model analysed differences in paralinguistic features such as pitch, pause, and others related to acoustic characteristics of the voice. This model has an accuracy rate of 86.4 per cent.

Previous research

In 2020, IBM Research collaborated with Pfizer to develop an AI tool to predict Alzheimer’s disease before its onset with a 71 percent accuracy rate. The study collected language samples from healthy people — who eventually did or did not develop Alzheimer’s symptoms. The tool demonstrated improved accuracy compared to prediction based on clinical scales (59 percent) and random choice method (50 percent).

The researchers considered over 87 variables in their study, including misspellings, punctuation, uppercasing, verbosity, lexical richness, and repetitiveness. In addition, they analysed language models to trace the distribution of word sequences. Beyond this, the researchers factored in the object naming, attention, abstraction, memory and test results from the Montreal Cognitive Assessment (MoCA).

IBM researchers then used NLP to analyse the participants’ language sample transcripts. The model picked up tiny subtleties and changes in discourses that are generally missed if done manually. Based on this, IBM researchers trained the ML model to account for multiple variables affecting the results. Lastly, they drew on data from the subjects at the Framingham Heart Study, where the participants are assessed through two-minute Mini-Mental State Examination speech tests every four years and neuropsychological exams every year. 

Fig. 3

CTT examples from FHS, including an unimpaired sample (a), an impaired sample showing telegraphic speech and lack of punctuation (b), and an even more impaired sample showing in addition significant misspellings and minimal grammatical complexity, e.g. lack of subjects (c).

Subscribe to our Newsletter

Get the latest updates and relevant offers by sharing your email.
Join our Telegram Group. Be part of an engaging community

Copyright Analytics India Magazine Pvt Ltd

Scroll To Top