基于不同全变差的医学图像压缩感知重构

Compressed sensing reconstruction of medical images based on different total variation

为提高磁共振成像(magnetic resonance imaging,MRI)的速度和成像质量,提出全变差(total variation,TV)、非局部全变差(nonlocal total variation,NLTV)和块稀疏全变差(group sparse total variation,GSTV)模型,对MRI图像压缩感知(compressed sensing,CS)重构。将不同TV模型分别与小波基相结合建立稀疏模型,由傅里叶矩阵进行测量,采用快速复合分裂算法(fast composite splitting algorithms,FCSA)实现MRI图像重构,以不同性能指标分析并比较不同TV模型的重构效果。实验结果表明,无论采样率如何设置,基于GSTV压缩感知重构MRI图像在性能指标以及细节精度等方面均具有明显优势,在快速医学成像领域具有一定临床应用价值。

To improve the clinical imaging speed and quality of magnetic resonance imaging （MRI） , total variation （TV） , nonlo-cal total variation （NLTV） and group sparse total variation （GSTV） were applied to compressed sensing （CS） reconstruction of MRI image. The different TV models were combined with wavelet transform for sparsity representation of the original image. The fast composite splitting algorithm （FCSA） was used to reconstruct MRI based on the Fourier matrix which was taken as the measurement matrix. Different indexes were applied to compare the reconstruction performances of the different TV models. Ex-perimental results demonstrate that GSTV is superior in terms of image quality indexes and detail precisions. The algorithm has certain clinical value in the field of rapid medical imaging.