AI Tool May Provide Faster, More Accurate Alzheimer's Diagnosis

Researchers at the Stevens Institute of Technology have developed an artificial intelligence (AI) tool that promises to accurately diagnose Alzheimer's disease without expensive scans or in-person testing, reports Technology Networks . The AI can not only diagnose the disorder at negligible cost with over 95 percent accuracy, but also can explain its conclusions so doctors can double-check accuracy. "We're . . . making it far easier to explain to patients why the AI came to the conclusion that it did, while diagnosing patients," said the institute's K.P. Subbalakshmi. She and her students designed an explainable AI engine that employs attention mechanisms and a convolutional neural network to create an algorithm that accurately identifies Alzheimer's telltales and detects subtle, previously missed linguistic indicators. They trained the algorithm using texts produced by healthy subjects and known Alzheimer's sufferers as they described a drawing of children stealing cookies from a jar. Subbalakshmi and her team then used tools from Google to render each individual sentence as a unique vector representing a specific point in a 512-dimensional space. By using those vectors in conjunction with handcrafted features, the AI was gradually able to spot similarities and differences between sentences spoken by healthy or unhealthy subjects, ascertaining the probability that any given text was to have been produced by someone with Alzheimer's.