计算机应用研究2023，Vol.40Issue(12)：3841-3847,7.DOI:10.19734/j.issn.1001-3695.2023.06.0227
基于特征尺度的平面波医学影像重建
Reconstruction of plane-wave medical image based on feature scale
摘要
Abstract
Compared with traditional line-scan imaging,plane-wave imaging is widely used due to its ultra-fast speed.How-ever,its poor imaging quality affects doctors'accurate diagnosis of tumors and vascular diseases.The existing techniques can improve the imaging quality but reduce the imaging frame rate,which cannot meet the demand for ultra-fast imaging in clinical medicine.To address the above problems,this paper proposed an image reconstruction method called generative adversarial net-work with multi scales and feature extraction(MF-GAN).Combined with multi-scale perceptual fields in the encoder,it used a U-Net-based generator to extract different levels of information.This paper proposed a fusion-sampling mechanism(FSM)in the decoder and combined it with cross-cross self-attention(CCSA)to extract local and global features.The MF-GAN was trained on the PICMUS 2016 dataset and used the combined loss to normalize the convergence direction.This model significant-ly improved reconstruction results in point targets,cyst targets,and in vivo authentic images compared to mainstream methods based on deep learning and beam synthesis.In summary,the MF-GAN model can solve the problem of unclear lesion sites in plane-wave images and reconstruct high-quality plane-wave images.关键词
平面波图像/多尺度/叠加采样机制/交叉自注意力Key words
plane-wave image/multiscale/fusion-sampling mechanism/cross-cross self-attention分类
信息技术与安全科学引用本文复制引用
杨翠云,侯钧译,曹怡亮,朱习军,闻卫军..基于特征尺度的平面波医学影像重建[J].计算机应用研究,2023,40(12):3841-3847,7.基金项目
山东省重点研发计划基金资助项目(2015GSF119016) （2015GSF119016）
青岛市科技惠民示范专项资助项目(23-2-8-smjk-20-nsh) （23-2-8-smjk-20-nsh）
山东省产教融合研究生联合培养示范基地项目(2020-19) （2020-19）
