Dictionary Based Low-Dose CT Image Processing

X-ray Computerized Tomography has the advantages of high spatial resolution, short scanning time, the capability of providing three-dimensional imaging and low patient cost. In recent years, CT scanning has been widely used in clinical diagnosis as an indispensable tool. However, with the growing popularity of CT technology, X-ray radiation during CT scan has gradually attracted people’s attention because the accumulation of radiation dose can cause cancer and other diseases. Dose can be reduced by adjusting tube current, but the resulting image will have more noise and artifacts which degrade the image quality and even lower diagnosis rate of lesions. So it is of great significance to obtain high quality CT images with low dose, and low dose CT (LDCT) is actually a hot research topic in the field of medical imaging. This thesis mainly focuses on discriminative dictionary based low dose CT image processing algorithm. The discriminative dictionary is then used to remove truncation artifacts. In the end, a novel method to remove metal artifacts is explored. Details are as follows:This thesis proposes an effective approach termed Discriminative Feature Representation (DFR) for LDCT image processing. This DFR method considers LDCT images as the superposition of high dose CT (HDCT) 3-D features and noise-artifact features (the collective noise and artifact features induced by low dose scan protocols), and the decomposed HDCT features can be used to yield the LDCT images with high quality. The proposed approach works through a discriminative representation using a featured dictionary composed of atoms to represent HDCT features and noise-artifact features. Boosted feature representations are brought by constructing dictionary using physical phantom images collected from the same CT scanner which generated the images to be processed. Allowing an easy implementation in practice, the proposed DFR method has good robustness in parameter setting for different CT scanner types. Comparative experiments with abdomen LDCT data validated the good performance of the proposed approach. Moreover, this algorithm has concise implementation and robust parameter settings, so it can be easily applied in conventional CT systems.In many cases, artifacts are more unacceptable in doctor’s diagnosis compared to noise. This thesis also studies reduction of two common artifacts in CT images. The idea of discriminative dictionary is used again to separate truncation artifacts in the image domain. Both experimental phantom studies and in vivo human subject studies were performed to validate the proposed method and to evaluate its performance. Finally, the reduction of metal artifacts was explored using consistency condition, and some preliminary results are given. The future research directions are discussed in the end.