4D cone-beam computed tomography for external-beam radiotherapy

Abstract

External Beam Radiation Therapy (EBRT) is one of the preferred cancer treatments, with precision and accuracy in delivering this radiation-type treatment is of paramount priority. Image-Guided Radiation Therapy (IGRT) has advanced over the years, which includes 3D Cone-Beam Computed Tomography (CBCT) as the standard imaging procedure for anatomical verification purposes. Studies have indicated several limitations in its current protocol. This includes truncation artifacts within the 3D reconstructed images that arise from the adopted operating mode for tumors located at large regions such as lungs. Slow-rotating CBCT image acquisition within one-minute for these regions also causes the images to be susceptible to motion-induced artifacts. Thus, 4D CBCT imaging has been opted as a means to consider internal motion during reconstruction. However, current 4D protocol requires additional equipment to estimate the motion and also has a longer acquisition period compared to 3D. There is also the general limitation of lack in ground truth data to validate any developed methods and algorithms. Therefore, the main objective of this thesis to address the aforementioned limitations is to analyze and develop an enhanced 4D CBCT imaging framework based on a typical 3D CBCT protocol. The adopted methodology is divided into three main phases. Firstly, the On-Board Imager (OBI) CBCT scanner that is currently installed at Universiti Kebangsaan Malaysia Medical Center is computationally modelled. 3D reconstruction algorithms that accommodate Half-Fan (HF) geometry are also developed in-house, which includes the comparison between two truncation artifact mitigation algorithms: projection weighting and complimentary rebinning. Simulations using these algorithms are compared with original images produced by the OBI. Twenty clinical datasets from two curative lung cancer patients are acquired within a three-month period. Concurrently, five sets of 4D CBCT digital phantoms are simulated using actual 4D Magnetic Resonance Imaging datasets based on a voxelized digital phantom development framework. Secondly, internal respiratory motion is estimated directly from the datasets that include irregular breathing patterns. Four main data-driven methods with three enhancement techniques are analyzed. The estimated signals are evaluated using a mean local correlation metric, ¯ 𝜌, against reference signals and are statistically analyzed, in which a more inclusive non-parametric Kruskal-Wallis comparison test is adopted. The results showed that Amsterdam Shroud with the proposed composite enhancement and Wiener filtering method displayed significantly accurate motion estimation ( ¯ 𝜌 = 0.89, 𝑃 < 0.01). Finally, 4D CBCT datasets are reconstructed using two algorithms and four binning strategies retrospectively. Evaluations are done using Tissue Interface Sharpness (𝜏) and Contrast-to-Noise Ratio (𝜍) metrics to measure motion and aliasing artifacts respectively. Both have significant correlations with the Root Mean Square Error (𝜉) metric (𝜌𝜉,𝜏 = 0.24, 𝑃 < 0.01, 𝜌𝜉,𝜍 = −0.47, 𝑃 < 0.01) for phantom datasets with available ground truth. The results showed that regardless of binning strategies and reconstruction algorithms, 4D CBCT reconstruction in general produces significantly sharper images compared to 3D ( ¯ 𝜏4D = 1.11 > ¯ 𝜏3D = 0.27, ¯ 𝑃 < 0.01). In conclusion, this thesis shows the advantages of 4D CBCT imaging using the standard 3D imaging protocol that could improve the accuracy of IGRT during EBRT treatments without prolonging the existing allocated time for each cancer patient.