Research on Optimization Scheme for Blocking Artifacts after Patch-Based Medical Image Reconstruction

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• High-resolution image processing with neural networks often involves slicing images into patches.
• Patch-based processing can lead to blocking artifacts due to zero-padding during convolution, impacting feature accuracy at patch edges.

Purpose of the Study:

• To develop and evaluate a novel fusion method for mitigating blocking artifacts in patch-stitched high-resolution images.
• To improve the quality of reconstructed images by addressing feature information loss at patch boundaries.

Main Methods:

• Proposed a fusion method assigning weights to pixels within image patches using a truncated Gaussian function.
• Transformed Euclidean distance from patch center into weight coefficients, decreasing with distance.
• Reconstructed images by applying these calculated weights.

Main Results:

• The proposed method effectively removed blocking artifacts in both simulated (BrainWeb) and real (Human Connectome Project) datasets.
• Evaluations using bias correction and denoising models demonstrated superior performance compared to five existing fusion methods.
• Achieved smoother bias fields and improved image quality.

Conclusions:

• The developed weighting-based fusion method is highly effective in eliminating blocking artifacts in neural network-processed high-resolution images.
• This approach offers a significant improvement over current state-of-the-art fusion techniques for image reconstruction.