Keio University Project Professor Taishiro Kishimoto and FRONTEO Co., Ltd. have developed a "Conversational Dementia Diagnosis Support AI Program" using natural language processing (NLP).AI can detect the possibility of dementia based on free conversation between the elderly and medical staff.
The number of people with dementia continues to increase in Japan, where the population is aging further, and it is estimated that by 2025, the number will reach 730 million (about one in five elderly people).Measures against dementia have become an important social issue in Japan.Diagnosis of dementia is usually made by medical history, imaging, and multiple cognitive tests to measure memory and arithmetic.However, these tests are highly specialized, requiring training for medical personnel who perform the tests, and are time consuming.
Dementia affects many functions, such as memory and attention, but it also affects language functions.Therefore, the research group recorded free conversations and conducted research to determine the possibility of dementia using natural language processing (NLP) technology.A total of 135 conversations were transcribed from 432 collaborators, decomposed into morphemes and parts of speech, vectorized, and machine-learned.As a result, we succeeded in correctly recognizing the presence or absence of dementia in a high percentage (accuracy 0.90, sensitivity 0.88, specificity 0.92).This accuracy could be achieved with the number of vocabularies obtained from an utterance of about 3 to 5 minutes.
The results of this research are based on the detection of dementia based on free conversations between the elderly and medical personnel, and it is possible to identify dementia without conducting tests such as memory and calculation.In the case of repeated examinations, the test subject may remember the contents of the examination and the accuracy of the examination may decrease, but this technology can avoid such a "learning effect", so it is expected to be put into practical use for screening examinations. and
Paper information:[Scientific Reports] Identifying neurocognitive disorder using vector representation of freeconversation