Iterative Reconstruction Technique for Cone-beam Computed Tomography with Limited Data

Abstract

X-ray cone-beam computed tomography (CBCT) has been extensively used in various applications, especially in medical analysis and the image-guided radiation therapy (IGRT). There have been on-going attempts to reconstruct images using a reduced number of projection data, in order to reduce the amount of radiation dose delivered to patients. However, reconstruction from insufficient number of projection data leads to reconstructed image with poor quality from severe artefacts when using analytical approach, i.e. Filtered Backprojection (FBP). In this scenario, iterative algorithms can significantly improve image quality, but comes at the cost of more complicated implementation and much longer computational time. These are the main drawbacks that make iterative algorithms difficult to be applied in a real clinical usage. This thesis focusses on developing advanced iterative algorithm to overcome the problems arising from CT reconstruction using limited number of projection data. The adaptive-weighted projection-controlled steepest descent (AwPCSD) algorithm is proposed by implementing projection onto convex sets (POCS) to enforce the data and the positivity constraints and minimising adaptive-weighted total variation (AwTV) norm. Experimental results showed that the AwPCSD algorithm is able to preserve the edges of the reconstructed image better with less number of sensitive hyper-parameters to tune, when compared to the pioneering work in this field such as the adaptive-steepest-descent POCS (ASD-POCS) algorithm. This thesis also analyses sensitivity of hyper-parameters, which are important components and play critical roles in the quality of reconstruction results. These hyper-parameters control the balance between the constraints and objective function in the TV regularised algorithms. The manual tuning of TV hyper-parameters is a tedious and time-consuming process, for which there is no well-established criteria to guarantee the optimal set of hyper-parameters for a given data apart from trials-and-errors. In order to overcome this problem, this thesis demonstrates 2 hyper-parameter selection approaches, which can be used to assist hyper-parameter selection from the user-defined ranges of hyper-parameters. The 2 algorithms employ 2 approaches, the Ant Colony Optimisation (ACO) and the Hedge, to select the best set of hyper-parameters for the implementation of the AwPCSD algorithm. Although the computational times for the training of hyper-parameters using these 2 algorithms are quite long, the set of hyper-parameters is guaranteed to produce a good quality of image, without having to manually re-select the values again. In addition, it is promising from the experimental results that the set of optimal hyper-parameters obtained from the training stage can also be applied to other datasets with the same imaging context. Thus, the time and resource spent on trying to figure out the best set of hyper-parameters for the best result of CT reconstruction using the TV regularised algorithms can be drastically saved, which eventually help to alleviate the complications of implementing iterative algorithms.

Date of Award	29 May 2019
Original language	English
Awarding Institution	University of Bath
Sponsors	Government of Thailand
Supervisor	Manuchehr Soleimani (Supervisor) & Ivan Astin (Supervisor)