Research On Low Dose4DCT Technique And Its Clinical Application

Respiration induces significant movements of thoracic and abdominal organs,which lead to motion artifacts on conventional CT images and distorted target volume,and degrade the efficiency and accuracy of radiotherapy. Four dimensional computedtomography (4DCT) could afford a set of CT images at different respiratory phases andthus account for respiratory motion, which helps to reduce motion artifacts and improvethe accuracy of radiatherapy. However,4DCT exposes high radiation dose to patients,which becomes a major concern and impedes its wide clinical application. Conventional4DCT techniques lead to obvious image noise and streak artifacts under low dosesituation. Therefore, we proposed two reconstruction algorithms for low dose4DCT inthe first half of this thesis. One is an iterative low dose CT slice reconstructionalgorithm based on edge-preserving total variation (EPTV) regularization, in which CTimages are reconstructed by minimizing an energy function consisting of a data fidelityterm and a regularization term. The EPTV term, proposed as the regularization term,introduces an adaptive penalty weight scheme to the original total variation (TV) term.By this penalty scheme, our algorithm preferentially performs smoothing on non-edgepixels to effectively remove image noise and artifacts due to low dose while betterpreserve low contrast edges. Another algorithm we proposed is an iterative low dose4DCT reconstruction algorithm based on temporal non-local means (TNLM), whichintroduces the temporal similarity between4DCT images at neighbouring phases intoreconstruction, so that different phases are reconstructed interrelatedly as opposed toindependently in conventional algorithms. In particular, the temporal regularizationis imposed between successive phases via TNLM term, which utilizes thesimilar image features at different phases to constructively enhance each otherand suppresses the image noise and streak artifacts by weighted average. Finally,we implemented these two iterative reconstruction algorithms on graphicsprocessing units (GPU) to speed them up by utilizing its parallel computingpower.However, there would still exist some problems when applying low dose4DCT techniques into clinic. First, the conventional4DCT technique depends on external respiratory monitors and thus could not be realized on conventionalCT scanners. While the available4DCT scanners are very expensive, whichimpedes wide application in developing countries. Second, the huge amount of4DCT images makes it a tedious and time-consuming job for physicians tomanually delineate the target and dangerous organs. To solve the first problem,we propose two strategies, virtual low dose4DCT method and a new4DCT sortingmethod based on image boundary discontinuity. The former only needs two imagesacquired at the end of exhale and the end of inhale by conventional CT scanners, andobtains a set of virtual4DCT images by image registration, interpolation anddeformation. This method not only supplies an alternative method to realize simulative4DCT function on conventional CT scanners, but also reduces radiation dose exposed topatients. The latter sorts CT images by minimizing the boundary discontinuitybetween neighboring couch positions. This method makes it feasible to realize4DCT on conventional CT scanners, and also avoids image artifacts due tosorting error when irregular respiratory happens. To solve the second problemreferred above, we proposed a semi-automatic4DCT delineation algorithmbased on texture feature of contour region. In this algorithm only the images atreference phase are required to be delineated manually, the contours on otherphases are automatically delineated by searching the regions with similar texturefeatures of contour regions. This method not only relieves physicians fromtedious target delineation, its repeatability also avoids accidental error due tomanual intervention.