Aug 28 2020
Scientists from the Stevens Institute of Technology have developed a new artificial intelligence (AI) algorithm that may lead to a precise diagnosis of Alzheimer’s disease, without the requirement for in-person testing or costly scans.
The team made this breakthrough by identifying slight variations in the way language is used by Alzheimer’s patients.
The AI software offers a more cost-effective method to diagnose Alzheimer’s disease with an accuracy of over 95%, and also has the potential to explain the conclusions relating to the disease, enabling doctors to double-check the precision of software’s diagnosis.
This is a real breakthrough. We’re opening an exciting new field of research, and making it far easier to explain to patients why the A.I. came to the conclusion that it did, while diagnosing patients. This addresses the important question of trustability of A.I .systems in the medical field.
K.P. Subbalakshmi, Founding Director, Stevens Institute of Artificial Intelligence, Stevens Institute of Technology
Subbalakshmi was responsible for developing the AI tool. She is also a professor of electrical and computer engineering at the Charles V. Schaefer School of Engineering & Science.
Scientists have known for a long time that Alzheimer’s disease can impact an individual’s use of language. People suffering from this disorder often use pronouns in the place of nouns, for example, saying “He sat on it” instead of “The boy sat on the chair.”
Alzheimer’s patients may also use uneasy circumlocutions, stating “My stomach feels bad because I haven’t eaten” rather than simply saying “I’m hungry.”
Subbalakshmi and her students designed an explainable AI engine that makes use of the convolutional neural network—a kind of AI that learns over time—and attention mechanisms—and ultimately developed software that can precisely detect the familiar telltale signs of Alzheimer’s disease and can also identify the slight linguistic patterns that were overlooked before.
Next, Subbalakshmi and her research team trained the AI algorithm by applying texts created by both familiar Alzheimer’s patients and healthy subjects as they explained a drawing of kids stealing cookies from a jar.
With the help of tools created by Google, Subbalakshmi and her group changed every single sentence into a special numerical vector, or sequence, indicating a particular point in a 512-dimensional space.
With such a technique, a concrete numerical value can be assigned to even complex sentences, rendering it easier to examine thematic and structural associations between sentences.
By applying those vectors together with handcrafted traits—those identified by subject matter experts—the AI system slowly learned to identify the differences and similarities between sentences articulated by unhealthy or healthy subjects, and was able to precisely determine how the given text could have been spoken by Alzheimer’s patients.
This is absolutely state-of-the-art. Our A.I. software is the most accurate diagnostic tool currently available while also being explainable.
K.P. Subbalakshmi, Founding Director, Stevens Institute of Artificial Intelligence, Stevens Institute of Technology
In association with her doctorate students, Mingxuan Chen and Ning Wang, Subbalakshmi presented the study results at the 19th International Workshop on Data Mining in Bioinformatics at BioKDD on August 24th, 2020.
In addition, the AI system can easily integrate new criteria that could be detected by other research groups in the days to come, which means it will only get more precise over time.
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.
K.P. Subbalakshmi, Founding Director, Stevens Institute of Artificial Intelligence, Stevens Institute of Technology
Theoretically, AI systems may someday diagnose Alzheimer’s disease based on any kind of text, right from a social media post or a personal email. But before that can be achieved, an algorithm had to be trained using several different kinds of texts created by familiar Alzheimer’s patients, instead of mere picture descriptions, and that kind of information is not available yet.
“The algorithm itself is incredibly powerful,” added Subbalakshmi. “We’re only constrained by the data available to us.”
In the next few months, Subbalakshmi expects to collect new information that will enable her AI software to be used for diagnosing patients on the basis of speech in languages, in addition to English. Subbalakshmi is also analyzing how other neurological disorders, such as depression, aphasia, traumatic brain injuries, and stroke—can impact the use of language.
“This method is definitely generalizable to other diseases. As we acquire more and better data, we’ll be able to create streamlined, accurate diagnostic tools for many other illnesses too,” concluded Subbalakshmi.