Exponential-Distance Weights for Reducing Grid-like Artifacts in Patch-Based Medical Image Registration

Area of Science:

• Medical Imaging
• Computer Vision
• Image Registration

Background:

• Patch-based medical image registration is widely used.
• Common methods generate grid-like artifacts due to zero-padding and sliding window extraction.
• These artifacts arise from uncertainty in feature extraction at patch edges.

Purpose of the Study:

• To propose a novel method for removing grid-like artifacts in patch-based medical image registration.
• To improve the accuracy and reliability of medical image registration.
• To introduce an exponential-distance-weighted (EDW) fusion technique.

Main Methods:

• Developed an exponential-distance-weighted (EDW) method for patch fusion.
• Utilized an exponential function to assign weights based on distance from the patch center.
• Lower weights were assigned to regions near patch edges to mitigate prediction uncertainty.

Main Results:

• The EDW method successfully removed grid-like artifacts.
• Achieved superior performance compared to several state-of-the-art methods on the OASIS-3 dataset.
• Demonstrated improved Dice Similarity Coefficient (DSC).

Conclusions:

• The proposed EDW patch fusion method effectively eliminates grid-like artifacts.
• This method enhances the accuracy of patch-based medical image registration.
• The EDW technique is compatible with existing patch-based registration models.