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

Abstract

Patch-based medical image registration has been well explored in recent decades. However, the patch fusion process can generate grid-like artifacts along the edge of patches for the following two reasons: firstly, in order to ensure the same size of input and output, zero-padding is used, which causes uncertainty in the edges of the output feature map during the feature extraction process; secondly, the sliding window extraction patch with different strides will result in different degrees of grid-like artifacts. In this paper, we propose an exponential-distance-weighted (EDW) method to remove grid-like artifacts. To consider the uncertainty of predictions near patch edges, we used an exponential function to convert the distance from the point in the overlapping regions to the center point of the patch into a weighting coefficient. This gave lower weights to areas near the patch edges, to decrease the uncertainty predictions. Finally, the dense displacement field was obtained by this EDW weighting method. We used the OASIS-3 dataset to evaluate the performance of our method. The experimental results show that the proposed EDW patch fusion method removed grid-like artifacts and improved the dice similarity coefficient superior to those of several state-of-the-art methods. The proposed fusion method can be used together with any patch-based registration model.

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

“…In the AAW method, the stride size has a great influence on the patch fusion effect. Therefore, on the HCP dataset, we verified the effect of different stride sizes on the fusion effect of AAW [ 5 ], EDW [ 9 ], and the method proposed in this paper. Table 3 lists the number of patches obtained on each image with different stride sizes and the fusion time of these three methods.…”

“…To evaluate the fusion ability of the proposed method after image reconstruction, we compared the proposed method with popularly available and state-of-the-art nonoverlapping patch splicing (NPS) method, arithmetic average weighted method (AAW) [ 5 ], MIScnn [ 7 ], Pathify [ 8 ], and Exponential-Distance-Weighted method (EDW) [ 9 ] on three datasets with different INU levels from BrainWeb. The dataset with INU = 20%and noise level = 3% was denoted as n3-20, the dataset with INU = 40%and noise level = 0% was denoted as n0-40, and the dataset with INU = 60%and noise level = 0% was denoted as n0-60.…”

“…Assuming that the pixel point Oði, jÞ overlaps m times, m = m i × m j , i ∈ ½0, M − 1, j ∈ ½0, N − 1, then m i can be calculated by Formula (6) [9], and the edge part that cannot be obtained by the patch is discarded. The value of m j can be obtained by replacing M in Formula (6) with N. Assuming that the m predicted values of the pixel point are P ði,jÞ…”

“…In the patchify method proposed by Wu, the overlapping region is covered with the next patch [ 8 ]. The EDW (Exponential-Distance-Weighted) method proposed by Wu et al [ 9 ] uses an exponential function to convert the distance between the point of the overlapping area and the center point of the patch into a weight coefficient and reconstruct the predicted deformation field by weighting. This method achieved good results.…”

“…Finally, the reconstructed image will be obtained by weighted fusion. For distance calculation, we choose Euclidean distance as the distance measure [ 9 ]. This method can be applied to any patch-based deep learning network without modifying the network structure and loss function.…”