Researchers from Stevens Institute of Technology designed an artificial intelligence software tool that can diagnose Alzheimer’s with 95% accuracy, reducing expensive diagnostic scans or in-person testing. The software program is also able to document and even explain its conclusions, so human experts can investigate the accuracy of its diagnosis.
Some tell-tale language signs the AI software looks to detect are how a person uses language and speech.
“This is a real breakthrough. We’re opening an exciting new field of research, and making it easier to explain to patients why the AI algorithm came to the conclusion that it did, while diagnosing patients. This is absolutely state-of-the-art.
Our AI software is the most accurate diagnostic tool currently available. This increases our ability to trust an AI system with important medical diagnosis” said K.P. Subbalakshmi PhD.
Alzheimer’s disease can affect a person’s use of language. By using AI software that learns over time, called a “convolutional neural network”, Subbalakshmi and her students developed a tool that accurately identifies well-known, tell-tale signs of Alzheimer’s, by detecting subtle language patterns that could easily be overlooked.
Subbalakshmi and her team trained their algorithm using text produced by both healthy subjects and known Alzheimer’s patients, as they described a drawing of children stealing cookies from a jar. Using tools developed by Google, Subbalakshmi and her team converted each sentence into a unique number sequence, called a vector, representing a specific point in a 512-dimensional space.
With this approach, complex sentences can be assigned a concrete number value. This makes it easier to analyze structural thematic relationships between sentences. By using those vectors along with hand-crafted features identified by subject matter experts, the AI software system gradually learned to spot similarities and differences between sentences spoken by healthy or unhealthy subjects. It can determine, with remarkable accuracy, the probability that a sample of speech belongs to an Alzheimer’s patient.
Also, the software can also easily incorporate new Alzheimer’s detection criteria that’s identified by other research teams in the future. So it will become more accurate over time.
The algorithm itself is incredibly powerful, we’re only constrained by the data available to us. We designed our system to be both modular and transparent. If other researchers identify new markers of Alzheimer’s, we can simply plug those into our architecture to generate even better results.
“This method can be used to detect other medical conditions. When we get more and better data, we’ll be able to create streamlined, accurate AI software tools to diagnosis many other illnesses too.”
The next step is to train the AI software on a much bigger volume of sample text. In the near future, AI software could diagnose Alzheimer’s based on any sample of text, from a personal e-mail, to a social-media post. To accomplish that goal, an algorithm needs to be trained on a large volume of sample texts, of different types, spoken or written by diagnosed Alzheimer’s patients. With a larger sample set containing the tell-tale language markers of Alzheimer’s disease, the software can become better familiar with what to look for.
Subbalakshmi is programming her software to diagnose patients using other languages. Her team is also exploring ways that other neurological medical conditions, like aphasia, stroke, traumatic brain injury, and depression, can affect a patient’s use of language.