Utilizing wavelet deep learning network to classify different states of task-fMRI for verifying activation regions.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Functional magnetic resonance imaging (fMRI) data is often noisy, complicating direct analysis of blood-oxygen-level dependent (BOLD) signals for intervention testing.
• Verifying brain activation regions identified through statistical analysis can be challenging due to data noise and complexity.

Purpose of the Study:

• To develop and validate a wavelet-based convolutional neural network (CNN) for verifying brain activation regions identified by statistical analysis.
• To assess the effectiveness of deep learning in distinguishing between activation and inactivation regions in fMRI data.

Main Methods:

• Utilized task fMRI data from public datasets (ds000157) involving visual stimuli (food vs. non-food pictures).
• Identified brain activation regions using statistical analysis with a P-value threshold of < 0.05.
• Designed a wavelet-based deep learning network to classify fMRI data from identified activation and inactivation regions.

Main Results:

• The CNN achieved 80.23% classification accuracy when using data from statistically identified activation regions.
• Classification accuracy dropped to 60% when using data from inactivation regions, highlighting the distinctiveness of activation areas.

Conclusions:

• The significant difference in classification accuracy confirms the validity and effectiveness of statistical methods for identifying brain activation regions.
• Deep learning and statistical analysis serve as complementary and cross-validating methods for human brain studies using fMRI data.