FNSAM: Image super-resolution using a feedback network with self-attention mechanism

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• High-resolution (HR) magnetic resonance imaging (MRI) is crucial for diagnosing brain lesions.
• Acquiring HR MRI images requires extended scan times and specialized equipment.

Purpose of the Study:

• To develop a deep learning-based super-resolution (SR) method for reconstructing HR MRI images from low-resolution (LR) inputs.
• To improve the efficiency and accessibility of obtaining detailed brain MRI data.

Main Methods:

• Proposed a novel feedback network with a self-attention mechanism (FNSAM) for SR reconstruction.
• Integrated a recurrent neural network (RNN) within the feedback loop to refine image features.
• Employed a self-attention mechanism (SAM) to extract critical hierarchical information for improved reconstruction.

Main Results:

• The FNSAM model demonstrated superior SR reconstruction performance compared to existing state-of-the-art methods.
• Quantitative evaluation using peak signal to noise ratio (PSNR) and structural similarity index measure (SSIM) confirmed the effectiveness of FNSAM.
• The method achieved more reasonable and accurate reconstructions of brain MRI images.

Conclusions:

• The proposed FNSAM method is effective for super-resolution reconstruction of MRI images.
• This approach offers a viable solution for generating high-quality MRI data efficiently.