Autism Diagnosis Using Transformer Architecture on Resting-State f-MRI
Parish, Connor, School of Engineering and Applied Science, University of Virginia
Feng, Xue, MD-BIOM Biomedical Eng, University of Virginia
Druzgal, Thomas, MD-RADL Neuroradiology, University of Virginia
Autism is a developmental disorder characterized by challenges with social skills, repetitive behaviors, speech, and nonverbal communication. The main method of diagnosis for autism is based off of a doctor’s analysis of an individual patient’s history and behaviors. Research is being done to try to use fMRI scans to diagnosis autism in order to improve diagnosis consistency and provide more information to improve treatment choices. The goal of this research is to gauge the efficacy of using a Transformer architecture to improve the automated diagnosis performance on the ABIDE I and ABIDE II datasets through the use of transfer and multi-task learning. The final model trained using a multi-task training method was able to achieve an accuracy of 0.689% on a subset of the ABIDE datasets. The most comparable previous work achieved an accuracy of 0.652% on the same subset of subjects.
BS (Bachelor of Science)
Autism, ABIDE, Transfomer, Sefl-Attention, Machine Learning, Transfer Learning, Multi-task
All rights reserved (no additional license for public reuse)