Current cone-beam Computed Tomography (CBCT) images have various shading artifacts that create challenges in accurately identifying tissue abnormalities and reduce their usefulness for clinical applications. The work performed in this thesis explores the efficacy of using planning CT (pCT) image as prior knowledge to improve CBCT image quality in radiation therapy (RT).
Since all RT patients routinely take multiple detector array CT (MDCT) image as part of the diagnostic procedure, the high quality MDCT serves as the “free” pCT. CBCT image is first registered with the pCT, and the primary CBCT projections are estimated via forward projections of the registered MDCT image. The low frequency errors in the projections, which are a major cause of artifacts in the reconstructed image, are estimated by filtering the difference between the estimated and raw CBCT projections. The corrected CBCT image is then reconstructed from the projections. With the pCT treated as ground truth, the CBCT image corrected by the proposed method is compared against the corrected image using the Varian algorithm, a commonly used industrial shading correction method. The results are presented in the axial, coronal, and saggital views, and are evaluated by comparing the errors of CT number, spatial non-uniformity (SNU) value, and image contrast value.
The proposed method is evaluated on 20 sets of patient data. The mean errors of CT number, SNU, and contrast value for the Varian corrected image and the image corrected by the proposed method in all three views are 53 HU and 41 HU, 7.3% and 3.0%, and 37 HU and 34 HU respectively. The results show that the proposed method delivers CBCT with better spatial uniformity and fewer CT number errors with 95% confidence, but with no significant variance in image contrast, than the Varian correction method.