MFAN: Multi-scale Feature Aggregation Network for Brain MRI Image Super-Resolution.

Area of Science:

• Medical imaging
• Artificial intelligence
• Neuroscience

Background:

• Magnetic resonance imaging (MRI) is crucial for diagnosing brain conditions.
• Current MRI super-resolution methods struggle with aggregating multi-scale textural details and high-frequency information.
• Accurate reconstruction is vital for reliable clinical diagnosis and application.

Purpose of the Study:

• To propose a novel Multi-scale Feature Aggregation Network (MFAN) for brain MRI image super-resolution.
• To address the challenge of effectively aggregating multi-scale features and high-frequency information for improved MRI reconstruction.
• To enhance the reliability of clinical diagnoses through advanced image super-resolution.

Main Methods:

• Developed a Multi-scale Feature Aggregation Network (MFAN) for brain MRI super-resolution.
• Incorporated Channel and Spatial Attention (CSA) mechanisms for shallow feature extraction.
• Introduced a Multi-scale Feature Aggregation Attention Block (MFAAB) for fusing diverse multi-pathway features.

Main Results:

• MFAN demonstrated superior performance compared to state-of-the-art methods on BraTS 2018 and Brain Tumor datasets.
• Achieved PSNR improvements of 1.054 dB and 0.609 dB at ×2 and ×4 magnifications on the BraTS 2018 dataset.
• Reported SSIM gains of 0.0128 and 0.0059 at ×2 and ×4 magnifications, respectively.

Conclusions:

• MFAN significantly advances brain MRI super-resolution, addressing key challenges in clinical neuroimaging.
• The network improves diagnostic precision by aggregating multiscale textural information and enhancing structural details.
• MFAN offers a potential solution for more reliable detection and diagnosis, reducing the need for repeated scans or high-field MRI systems.