Predicting Autism From MRI

Predicting the Diagnosis of Autism using Classification Models based on fMRI

Abstract

Autism spectrum disorder is a lifelong neurodevelopmental disorder that is diagnosed based on behavioral and social interaction patterns. Predictive algorithms provide a novel approach in identifying key neurological biomarkers and subsequent psychiatric diagnosis using functional magnetic resonance imaging (fMRI) data. This study analyzes a dataset collected from studies at 17 international locations as part of the Autism Brain Imaging Dataset Exchange (ABIDE). To classify the medical images, we extract regional homogeneity scores and compare random forest models with convolutional neural networks. The accuracy for a five-fold cross validated random forest model and neural network having the highest validation score were found to be 0.57 and 0.54 respectively using the holdout test dataset.

See my report and Jupyter notebook for further details.

References

Martino, A. di, C. G. Yan, Q. Li, E. Denio, F. X. Castellanos, K. Alaerts, J. S. Anderson, et al. 2014. “The Autism Brain Imaging Data Exchange: Towards Large-Scale of the Intrinsic Brain Architecture in Autism.” Molecular Psychiatry 19 (6): 659. https://doi.org/10.1038/MP.2013.78.

Miller, Tom. “Medical Imaging”. MSDS 422: Practical Machine Learning. Course at Northwestern University, Chicago, IL, May 21, 2022.