Using a generative computer modeling technique, a multi-university team has developed a method to identify genetic markers of autism from brain images with 89% to 95% accuracy (the article does not explain the range). Currently, autism is diagnosed based on its complex behavioral presentation. What’s particularly interesting about the algorithm—beyond the identification of non-behavioral markers—is how it tracks molecular movement to produce a new visualization and model for analysis.
Most machine learning approaches to image pattern analysis focus on single-pattern recognition. This is useful for identifying positive cases but is even more valuable for negatives, as it helps rule out patterns that don’t require further diagnostic attention. However, this new algorithm compiles images to create an entirely new dataset.
UVA Research Cracks the Autism Code, Making the Neurodivergent Brain Visible
|