Regression-based machine-learning approaches to predict task activation using resting-state fMRI.

Area of Science:

• Neuroimaging
• Machine Learning
• Computational Neuroscience

Background:

• Resting-state functional magnetic resonance imaging (fMRI) is increasingly used to predict brain activity during tasks.
• The general linear model (GLM) is a common but potentially simplistic method for this prediction.

Purpose of the Study:

• To compare the predictive performance of various machine learning regression models against the GLM for resting-state fMRI prediction of task activation.
• To evaluate the impact of training dataset size on prediction accuracy.

Main Methods:

• Analysis of resting-state and task-based fMRI data from 350 Human Connectome Project subjects.
• Comparison of prediction accuracy and Dice coefficient across GLMs, feed-forward neural networks, and random forest bagging.
• Evaluation of the effect of training sample size on predictive performance.

Main Results:

• Prediction accuracy improved with increased training data, showing an initial steep rise before plateauing around 30-40 subjects.
• All models demonstrated strong performance in predicting task activation, indicated by correlation matrices with a clear diagonal trend.
• Neural networks and random forest bagging generally outperformed the GLM in prediction accuracy.

Conclusions:

• While GLMs offer a functional approach, more complex machine learning techniques like neural networks and random forest bagging show superior performance for predicting task activation from resting-state fMRI.
• The enhanced computational demands of these advanced models are a trade-off for improved prediction accuracy in neuroimaging analysis.