Diagnosing ADHD from fMRI Scans Using Hidden Markov Models

TL;DR

This study employs hidden Markov models on fMRI data to diagnose ADHD, analyzing brain activity patterns over time with various dimensionality reduction techniques, achieving around 62-63% accuracy.

Contribution

It introduces a novel application of hidden Markov models combined with PCA and kernel PCA for ADHD diagnosis from fMRI scans.

Findings

Achieved approximately 62-63% accuracy in ADHD classification.

Demonstrated effectiveness of PCA and kernel PCA in dimensionality reduction for this task.

Applied the model to the ADHD-200 dataset with promising results.

Abstract

This paper applies a hidden Markov model to the problem of Attention Deficit Hyperactivity Disorder (ADHD) diagnosis from resting-state functional Magnetic Resonance Image (fMRI) scans of subjects. The proposed model considers the temporal evolution of fMRI voxel activations in the cortex, cingulate gyrus, and thalamus regions of the brain in order to make a diagnosis. Four feature dimen- sionality reduction methods are applied to the fMRI scan: voxel means, voxel weighted means, principal components analysis, and kernel principal components analysis. Using principal components analysis and kernel principal components analysis for dimensionality reduction, the proposed algorithm yielded an accu- racy of 63.01% and 62.06%, respectively, on the ADHD-200 competition dataset when differentiating between healthy control, ADHD innattentive, and ADHD combined types.