Brain Decoding Using fMRI Images for Multiple Subjects through Deep Learning.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Functional Magnetic Resonance Imaging (fMRI) noninvasively decodes human cerebral conditions by exploring neuronal activity.
• Deep learning models can interpret complex brain patterns from fMRI data, but high dimensionality and training time pose challenges.
• Existing methods struggle with efficient learning of high-level features and accurate voxel interpretation, leading to misclassification errors.

Purpose of the Study:

• To propose an improved Convolutional Neural Network (CNN) for enhanced analysis of fMRI data.
• To address challenges of high dimensionality and prolonged training times in deep learning models for brain activity decoding.
• To improve the accuracy and efficiency of classifying human thoughts from fMRI imaging data.

Main Methods:

• Implemented an improved CNN with functionally aligned features and dimensionality reduction.
• Utilized autoadjusted weights and optimal activation functions for transforming high-dimensional feature vectors into a low-dimensional space.
• Incorporated the Swish activation function to enhance model density and reduce training time.

Main Results:

• The proposed CNN model demonstrated superior performance compared to existing classifiers.
• Experimental results showed significant improvements in accuracy for brain voxel interpretation.
• The model achieved efficient learning and accurate classification of thoughts from fMRI data within reduced training times.

Conclusions:

• The developed CNN technique offers a more efficient and accurate approach for decoding brain activity from fMRI data.
• Functional alignment and dimensionality reduction are key to improving deep learning model performance in neuroscience applications.
• The Swish activation function contributes to faster and more efficient training of deep learning models for complex neuroimaging analysis.