Comprehensive processing method for Ghost artifacts in low-field diffusion-weighted imaging based on attention residual UNet and accelerated non-mean filtering
XU Yang
WEI Jing
KIM Siseung
ZHANG Huiyao
LI Bingkeong
Abstract：To address the N/2 Ghost artifacts caused by phase encoding errors in diffusion-weighted imaging(DWI)for low-field(<1 T)magnetic resonance imaging(MRI)systems,we proposed an approach integrating deep learning and optimized filtering to e-liminate Ghost artifacts and enhance image quality.Firstly,an AR-UNet model incorporating dense residual connection and attention gate mechanism was developed to achieve precise segmentation of craniocerebral anatomical structures through feature reuse and dynam-ic weight allocation.Then,an edge-constrained accelerated non-local means filtering(SCNLM)was used to improve the computation-al efficiency of the model.The results showed that the average Dice similarity coefficient,accuracy rate and specificity of the model reached 0.932 1,0.943 6 and 0.943 0,respectively.SCNLM could increase the computational efficiency of the traditional NLM algo-rithm by approximately 50%,while maintaining the peak signal-to-noise ratio of 29.50 dB and the structural similarity of 0.88.This research can effectively suppress Ghost artifacts in low-field MRI systems and significantly enhances image quality.
Keywords：Low-field MRIDiffusion-weighted imagingGhost artifactsAttention residual UNetAccelerated non-local means filtering
Publication Date：2025-06-30
Online Publishing Date：2026-09-11(First online date of this platform, not the publication date of the document)
Pages：8( 162-169 )