基于稀疏表示和字典学习的低剂量CT图像恢复研究

发布时间：2018-08-06 13:11

【摘要】：随着计算机技术的不断发展,CT成像技术已经广泛运用到临床疾病诊疗中,成为了检查脑疾病诊断的首选。但是CT成像扫描时X射线剂量的使用量也随之增加,使得受检人员诱发疾病几率增大,而低剂量CT可降低它诱发疾病的几率。但是低剂量的同时引起了图像质量的下降,因此对低剂量CT图像进行恢复研究具有重大意义。由于稀疏表示和字典学习因其优良特性而被应用于图像去噪、恢复等信号问题的求解,因此将该方法应用于低剂量CT图像问题具有重要意义和研究价值。本文为解决脑低剂量CT图像降质问题,对其恢复研究做了如下工作:首先,从基于字典学习(MOD、K-SVD、OLM、FDL-PG)的稀疏表示进行低剂量CT图像恢复研究,结果表明FDL-PG算法相对其它算法在视觉感受和客观上都更优,且适应性好、收敛速度快,但仍存有一些噪声和部分信息缺失等问题。然后,针对上述问题,本文提出了两种改进的基于稀疏表示和字典学习的低剂量CT图像恢复方法。一种方法是先对低剂量体模和临床脑CT图像进行主成分分析(PCA),再用降维后的数据进行字典训练(FDL-PG)并去噪。该方法(FDL-PG-PCA)增强了去噪性能,但仍有少量细节问题丢失。另一种方法是先对低剂量体模和临床脑CT图像进行BM3D滤波处理,再用滤波后的数据进行字典训练(FDL-PG)并去噪,该方法(FDL-PG-BM3D)对细节信息保持良好。通过对体模和临床CT图像实验结果表明这两种方法去噪性能较高,进一步抑制了噪声。这两种方法在临床运用中,有望在保证医生诊断准确的同时减少受检患者接收X射线辐射剂量。
[Abstract]:With the development of computer technology, CT imaging technology has been widely used in the diagnosis and treatment of clinical diseases, and has become the first choice for the diagnosis of brain diseases. However, the amount of X-ray dose used in CT scan also increases, which increases the probability of inducing disease, while low dose CT can reduce the probability of inducing disease. However, low dose leads to the deterioration of image quality, so it is of great significance to study the restoration of low dose CT images. Since sparse representation and dictionary learning are used to solve signal problems such as image denoising and restoring due to their excellent characteristics, it is of great significance and research value to apply this method to low-dose CT image problems. In order to solve the problem of brain low-dose CT image degradation, the following work has been done in this paper: firstly, low-dose CT image restoration is studied from sparse representation based on dictionary learning (MODK-SVDU OLMM-FDL-PG). The results show that the FDL-PG algorithm is better than other algorithms in visual perception and objectivity, and has good adaptability and fast convergence speed, but there are still some problems such as noise and lack of some information. Then, two improved low dose CT image restoration methods based on sparse representation and dictionary learning are proposed. One method is to perform principal component analysis (PCA),) on low dose phantom and clinical brain CT images, then to do dictionary training (FDL-PG) and denoising with dimensionally reduced data. This method (FDL-PG-PCA) improves the denoising performance, but there are still a few details lost. Another method is to process low dose phantom and clinical CT images with BM3D filter, then use filtered data for dictionary training (FDL-PG) and denoising. This method (FDL-PG-BM3D) maintains good detail information. The experimental results of phantom and CT images show that these two methods have high denoising performance and further suppress the noise. These two methods are expected to ensure the accuracy of the diagnosis and reduce the dose of X-ray radiation to the patients.
【学位授予单位】：东华理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】