Research on the Modality Transfer Method of Brain Imaging Based on Generative Adversarial Network.

Area of Science:

• Neuroimaging
• Medical Technology
• Artificial Intelligence

Background:

• Functional magnetic resonance imaging (fMRI) provides detailed anatomical insights but is costly and unsuitable for patients with metal implants.
• Electroencephalography (EEG) is a more accessible brain imaging technique, yet it lacks the spatial resolution of fMRI.
• Bridging the gap between EEG and fMRI is crucial for broader application in studying brain diseases.

Purpose of the Study:

• To develop a novel brain imaging modality transfer framework, BMT-GAN, to convert EEG data into fMRI data.
• To overcome the limitations of fMRI, such as high cost and contraindications for certain patients.
• To provide comprehensive reference information for radiologists by enabling EEG-to-fMRI data conversion.

Main Methods:

• Proposed a generative adversarial network (GAN)-based framework named BMT-GAN for brain imaging modality transfer.
• Introduced a novel non-adversarial loss function to minimize perceptual and style discrepancies between input (EEG) and output (fMRI) images.
• Validated the framework's ability to convert EEG modality data to fMRI modality data.

Main Results:

• Successfully demonstrated the conversion of EEG data to fMRI modality data using the BMT-GAN framework.
• The proposed non-adversarial loss effectively reduced differences between the modalities.
• Qualitative and quantitative comparisons showed the superiority of BMT-GAN over existing GAN-based transfer methods.

Conclusions:

• BMT-GAN offers a promising solution for brain imaging modality transfer, enhancing the utility of EEG data.
• The framework provides a cost-effective and accessible alternative to traditional fMRI for brain disease research.
• This approach has the potential to improve diagnostic capabilities and provide valuable insights for radiologists.