A new study suggests that technology similar to ChatGPT has the ability to detect early signs of Alzheimer’s disease.
The study, led by Drexel University’s School of Biomedical Engineering, Science and Health Systems, revealed that GPT-3 — which is a language model more powerful than ChatGPT — can accurately diagnose dementia 80% of the time. GPT-3 was developed by OpenAI, the same company that created ChatGPT.
Hualou Liang, a Drexel professor and co-author of the study, shared his thought process in exploring the benefits of ChatGPT in medical research.
“We know everyone’s been talking about AI in particular after GPT came out [and] even before that,” Liang said. “As someone involved in biomedical engineering, it’s quite natural for me to think about what I can do to solve biomedical and health care problems.”
The study, published on Dec. 22, relied on analyzing individuals’ speech when asked questions and to perform a series of tasks. In collaboration with the United States National Institutes of Health, researchers trained the AI program to read transcripts and analyze speech patterns in the transcript.
The knowledge encoded in the GPT-3 model is used to generate a vector representation of a transcription from speech. Aside from distinguishing individuals with Alzheimer’s from healthy controls, GPT-3 can infer a person’s cognitive test score based on speech data.
Programs like GPT-3 can detect pronunciation mistakes, hesitation, missed comprehension, and grammar and punctuation mistakes. Language impairment is a common biomarker for neurodegenerative disorders like Alzheimer’s, affecting between 60 and 80% of dementia patients. Therefore, GPT-3 could help determine if a patient should take the next step of diagnosis and examination.
“GPT-3 can pick up many subtle differences [in speech]. This could be the reason why this mode performs remarkably well,” Liang said.
Liang elaborated on how artificial intelligence could decipher language patterns and help identify potential Alzheimer’s patients.
“When compared to a healthy control, people with [Alzheimer’s] have a lot of repetition, have incomplete thoughts, and sometimes forget the task,” Liang said. “They may make a grammar or pronunciation mistake and may have long pauses and filler words.”
With a growing dataset of interviews including Alzheimer’s patients, researchers are hopeful that identifying future patients could become easier with AI. The results suggest that AI could become an accessible and generally reliable tool for community-based Alzheimer’s assessment.
The Drexel researchers are currently in talks with Penn to develop an app or a web application using AI for dementia detection, according to Liang.
“Everyone has a cell phone so, with this app, people may not need to make a trip to the doctor or clinic. ... There is a huge potential and growing market for this particular application,” Liang said.