Predicting brain age using Tri-UNet and various MRI scale features.

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Human aging involves significant brain tissue changes.
• Accurate brain age prediction is crucial for disease risk screening and diagnosis.
• Existing methods face challenges in learning sufficient features from neuroimaging data.

Purpose of the Study:

• To develop an accurate method for predicting brain age.
• To address limitations in learning image features from brain neuroimaging data.
• To improve the assessment of age-related brain changes and disease risks.

Main Methods:

• Proposed a multi-scale feature fusion method named Tri-UNet, based on the U-Net architecture.
• Implemented a brain region information fusion method utilizing multi-channel input networks.
• Leveraged features at different scales from MRI scans and integrated information from various brain regions.

Main Results:

• Achieved a minimum Mean Absolute Error (MAE) of 7.46 on the Cam-CAN dataset.
• Demonstrated effective utilization of multi-scale and multi-region features from MRI data.
• Validated the proposed method's efficacy in brain age prediction.

Conclusions:

• The Tri-UNet method offers a novel approach to feature learning for brain age prediction.
• This advancement contributes to better understanding and managing age-related neurological conditions.
• The findings have significant implications for practical applications, including elderly care and education.