Study suggests ChatGPT could be a screening tool for Alzheimer’s disease

ChatGPT, a now-famous artificial intelligence (AI) program, has made headlines in recent months for its ability to build human-like text responses. A new report suggests the tool could also use its language analysis skills to help diagnose people with early signs of Alzheimer’s disease.

in new research was announced in PLOS Digital Healthresearchers at Drexel University in Philadelphia tested the AI ​​program ChatGPT to see if it could pick up on subtle language cues associated with Alzheimer’s disease.

ChatGPT is a language model developed by research firm OpenAI. The initials in its name stand for General Pretrained Transformer. The program has been trained with vast amounts of text and information to answer questions and produce text in an easy-to-read format that reflects the grammar and syntax of human writing.The preview version of the software is Open to the public in November.

ChatGPT has garnered attention for its ability to generate text that appears indistinguishable from human-written text, but a new study evaluates ChatGPT’s ability to do the opposite. That is, it looks at transcriptions of human speech to identify speech patterns and usage that appear unnatural.

LianghuaroDrexel professor of biomedical engineering, Ph.D., and PhD student Feli Agvaor explained that one way clinicians can identify early-onset Alzheimer’s disease is by analyzing voice. . Approximately 60-80% of people with Alzheimer’s disease have symptoms of language impairment.

“The tests most commonly used for early detection of Alzheimer’s disease look at acoustic features such as pauses, intelligibility and voice quality in addition to cognitive tests,” Liang said in a press release. , we believe that improvements in natural language processing programs offer another avenue to support early detection of Alzheimer’s disease.”

Researchers used transcriptions of Alzheimer’s patients’ speech to create a characteristic profile of Alzheimer’s patients’ speech. This profile is sensitive to the nuances of how Alzheimer’s patients structure sentences, use words, and sometimes forget the meaning of words. The author then used that profile to train her ChatGPT-3 and screen her text samples for signs of Alzheimer’s disease. The goal of the program was to distinguish between transcripts of Alzheimer’s patients’ speech and those of normal subjects.

Using a training profile created for this study, Liang and Agbavor found that ChatGPT can accurately predict Alzheimer’s disease about 80% of the time. Others analyzed outperformed two natural language processing programs, they said.

The researchers next decided to see if this type of analysis could be used to predict patient performance in other Alzheimer’s disease assessments. I requested. mini mental state exam (MMSE), a series of commonly used questions used to assess the severity of dementia in patients. They found that ChatGPT was approximately 20% more accurate in predicting his MMSE score than conventional assessments based on acoustic features.

Liang said the research suggests that ChatGPT could be a promising diagnostic tool. He and his Agbavor also highlighted some important caveats. First, he noted that speaking patterns differ across regions and demographic groups, so diagnostic tools are based on data from around the world, and it is important to assess bias. Patient privacy must be protected. We also need to increase our confidence in AI before it can be widely used in healthcare applications. Especially since machine learning programs can make predictions using super-complex “reasons” that even human developers can’t explain.

Still, Liang said the study shows that tools like ChatGPT could be a meaningful addition to healthcare systems if these issues can be addressed.

“Our proof of concept indicates that this could be a simple, accessible, and sufficiently sensitive tool for community-based testing, which could be used for early screening and risk assessment prior to clinical diagnosis. It can be very useful for evaluation.”