The Next Mark for GPT-3: Diagnosing Alzheimer’s Disease by Voice

Alzheimer’s disease has one seemingly simple early sign that is not often talked about. It’s a subtle change in speech patterns.

Increased hesitation. Grammar mistake. You forget the meaning of words, or you mispronounce common words or favorite phrases or idioms that used to flow naturally.

Scientists have long sought to decipher this language degeneration as an early indicator of Alzheimer’s disease. One idea is to use natural language software as a kind of “guide” to track anomalous use of language.

Simple, isn’t it? Here’s the problem. Everyone speaks differently. It seems obvious, but this is a big headache for AI. Our speech patterns, rhythms, tones, and word choices are all colored with shades of personal histories and nuances that the average language AI struggles to decipher. A statement that is ironic to one person may be perfectly sincere to another. Repeated grammatical errors are personal habits from decades of misuse that are now difficult to change or may be a reflection of dementia.

So why not take advantage of the most creative AI language tools available today?

in a study was announced in PLOS Digital Healtha team at Drexel University has taken a major step forward in bridging the creative power of GPT-3 with neurological diagnostics. Using this, the team retrained GPT-3 to select linguistic nuances suggestive of dementia.

When fed with new data, the algorithm was able to reliably detect Alzheimer’s disease patients from healthy patients and predict their cognitive test scores. No additional knowledge of the patient or their medical history was required.

“To our knowledge, this is the first application of GPT-3 to predict dementia from speech,” said the authors. “Using voice as a biomarker enables rapid, inexpensive, accurate and non-invasive AD diagnosis and clinical screening.”

early bird

Despite scientific best efforts, diagnosing Alzheimer’s disease is extremely difficult. Due to its often genetic nature, this disorder has no unified theory or treatment. But what we do know is that areas in the brain associated with memory begin to accumulate clumps of proteins that are toxic to neurons. It accelerates your decline and eventually erodes everything that makes you up.

The most dangerous part of Alzheimer’s disease is that it’s difficult to diagnose. allows early capture of these proteins. But scientists have long known that cognitive symptoms can creep in long before the protein clumps appear.

There is a ray of hope here. Even without a cure, diagnosing Alzheimer’s disease early can help patients and their families develop plans for support, mental health, and finding treatments to manage symptoms. With recent approval, this groupis a drug that may help moderately protect against cognitive decline in patients with early-stage Alzheimer’s disease, and the race to detect the disease early is heating up.

Speak Your Mind

Instead of focusing on brain scans and blood biomarkers, Drexel’s team turned to voice, a much easier method.

“Ongoing research shows that the cognitive effects of Alzheimer’s disease can manifest itself in language production.” Said Study author Dr. Hualou Liang. “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.”

This idea has long been pursued by cognitive neuroscientists and AI scientists. Natural language processing (NLP) has dominated his AI field in its ability to recognize everyday language. By inputting patient voice or written recordings, neuroscientists can emphasize specific vocal “tics” in certain groups of people (for example, people with Alzheimer’s disease).

Sounds great, but these are highly coordinated studies. They rely on knowledge of specific issues rather than more general Q&A. The resulting algorithms are handcrafted and difficult to scale to a wider population. It’s like going to a tailor for a suit or dress that fits perfectly, only to realize after a few months that it doesn’t fit other people or even yourself.

It’s a diagnostic issue. Alzheimer’s disease, or other neurological disorders, tend to progress. Algorithms trained in this way “make it difficult to generalize to other stages of progression and disease types that may correspond to different linguistic features,” the authors say.

In contrast, GPT-3’s underlying Large Language Model (LLM) is much more flexible, offering “powerful universal language understanding and generation”.

One particular aspect caught their eye: embedding. Simply put, it means that an algorithm can learn from large amounts of information and generate some kind of ‘idea’ for each ‘memory’. Using this trick on texts can reveal additional patterns and features beyond what most trained professionals can detect, the authors say. In other words, GPT-3-powered programs based on text embeddings may be able to detect differences in speech patterns that escape neurologists.

“A systematic approach to language analysis and generation of GPT-3 makes it a promising candidate for identifying subtle speech features that may predict the development of dementia.” Said Study author Felix Agboh. “Using a huge dataset of interviews, he trained GPT-3, some of which was from patients with Alzheimer’s disease, and provided the information needed to extract speech patterns that could be used in future patients. It can be applied to identify markers.”

creative solutions

The team immediately used GPT-3 for two important measures of Alzheimer’s disease. distinguishing Alzheimer’s patients from healthy patients and predicting the severity of dementia in patients based on a cognitive benchmark called the Mini-Mental State Exam (MMSE).

Like most deep learning models, GPT-3 is very data hungry.where the team ADReSSo Challenge (Alzheimer’s Dementia Recognition through Spontaneous Speech) contains everyday conversations of people with and without Alzheimer’s disease.

For the first challenge, the team compared the GPT-3 program to two programs that track specific “tics” of the language. both models, Ada When Babbage (in honor of the computing pioneer) far outperformed traditional models based solely on acoustic capabilities. The algorithm did even better when predicting the accuracy of the dementia MMSE on speech features alone.

Compared to other state-of-the-art Alzheimer’s disease detection models, the Babbage version beats the competition in accuracy and recall.

“All these results suggest that GPT-3-based text embeddings are a promising approach for AD assessment and may improve early diagnosis of dementia,” the authors said. .

With all the hype around GPT-3 and AI in healthcare in general, it’s easy to lose sight of what really matters: patient health and well-being. Alzheimer’s disease is a terrible disease that literally erodes the mind. Early diagnosis is information, and information is power. This helps inform life choices and evaluate treatment options.

“Our proof of concept shows that this could be a simple, accessible, and sufficiently confidential tool for community-based testing,” said Liang. “This could be very useful for early screening and risk assessment before clinical diagnosis.”

Image credit: NIH (National Institutes of Health)