基于剪切波变换的下卷积正则化SPECT图像重建

Infimal Convolution Regularized SPECT Image Reconstruction Based on Shearlet Transform

摘要: 单光子放射断层成像(SPECT)在疾病的影像诊断中起重要作用。全变差正则项在抑制噪声方面非常有效，但会产生阶梯状伪影。剪切波变换是一种多尺度几何分析方法，它比传统的小波变换更符合人类视觉系统的感知特性，能更有效地刻画和捕获图像中的边缘、纹理等几何特征，并能充分利用图像自身的几何特性实现对其更为“稀疏”的表示。本文提出一种基于下卷积剪切波变换惩罚项的SPECT重建方法，在抑制噪声的同时，进一步消除阶梯状伪影。实验结果表明，本文方法优于下卷积全变差惩罚项，图像边缘清晰、细节丰富，符合人眼视觉效果。

Abstract: Single-photon emission tomography (SPECT) plays an important role in the imaging diagnosis of the disease. Total variation regularization is more effective in suppressing noise but produces stair-step artifacts. Shearlet transformation is a multi-scale geometric analysis method. It is more in line with the traditional wavelet transform perception characteristics of the human visual system, and can more effectively depict and capture geometric feature texture in the image edge. It can make full use of the geometric characteristics of the image itself to its more sparse representation. In this paper, Infimal Convolution regularized SPECT image reconstruction based on shearlet transform is proposed, suppressing the noise and further eliminating stair-step artifact. Results show that the method is better than the infimal convolution total variation punishment. The reconstructed image has a clear edge and rich detail, which conforms to the human visual effect.