Advancing Metaverse-Based Healthcare With Multimodal Neuroimaging Fusion via Multi-Task Adversarial Variational...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• The metaverse integrates physical and virtual realities, offering new healthcare applications.
• AI-driven medical imaging, particularly brain age estimation, is crucial for detecting neuropathologies like Alzheimer's disease.
• Existing deep learning methods struggle with multimodal functional MRI (fMRI) data for brain age prediction.

Purpose of the Study:

• To develop a novel deep learning framework, the Multitask Adversarial Variational Autoencoder (M-AVAE), for enhanced brain age prediction.
• To integrate structural MRI (sMRI) and fMRI data effectively for more accurate brain age estimation.
• To address challenges in handling complex fMRI data structures and noise in functional connectivity.

Main Methods:

• Developed the M-AVAE, a deep learning framework that separates latent variables into generic and unique codes.
• Integrated multitask learning with sex classification to account for sex-specific aging patterns.
• Evaluated the M-AVAE on the OpenBHB multisite brain MRI dataset.

Main Results:

• The M-AVAE achieved a mean absolute error of 2.77 years in brain age prediction.
• Demonstrated superior performance compared to conventional methodologies.
• Successfully integrated multimodal MRI data, isolating shared and modality-specific features.

Conclusions:

• The M-AVAE framework significantly improves brain age estimation accuracy using multimodal MRI data.
• This approach holds promise for metaverse-based healthcare applications, particularly in early detection of age-related brain conditions.
• The developed framework offers a powerful tool for advancing AI in neuroimaging analysis.