Accurate predictions of individual differences in task-evoked brain activity from resting-state fMRI using a sparse...

Area of Science:

• Neuroscience
• Neuroimaging
• Machine Learning

Background:

• Predicting individual differences in task-fMRI activity has broad applications in neuroscience.
• Models using resting-state fMRI data have shown promise for high predictive accuracy.

Purpose of the Study:

• To propose and evaluate improvements to predictive models for individual differences in task-fMRI activity.
• To enhance the accuracy and utility of models that predict brain activity patterns.

Main Methods:

• Utilized a sparse ensemble learner for improved feature extraction.
• Employed Stochastic Probabilistic Functional Modes (sPROFUMO) for feature extraction, outperforming the dual-regression approach.
• Developed a model capable of separately predicting the shape and intensity of individual task activations.
• Incorporated training on residual differences in brain activity to further boost predictions.

Main Results:

• Features extracted using sPROFUMO demonstrated superior performance compared to the dual-regression method.
• The model successfully modeled shape and intensity of individual task activations distinctly.
• Training on residual differences significantly enhanced individual prediction accuracy.
• The proposed model achieved state-of-the-art prediction accuracy, comparable to task-fMRI test-retest reliability.
• Results were validated on both Human Connectome Project (surface-based) and UK-biobank (volumetric) data.

Conclusions:

• The novel modeling approach significantly improves the prediction of individual differences in task-fMRI.
• The method offers a potential supplement to traditional task localizers in neuroscience.
• The findings highlight the potential of resting-state fMRI for predicting task-based brain activity patterns.