CBCT image enhancement for improving accuracy of radiomics analysis and soft tissue target localization

Cone-beam computed tomography (CBCT) is one of the most commonly used image modalities in radiation therapy. It provides valuable information for target localization and outcome prediction throughout treatment courses. However, CBCT images suffer from various artifacts caused by scattering, beam hardening, undersampling, system hardware instability, and motions of the patient, which severely degrade the CBCT image quality. In addition, CBCT images have extremely poor soft-tissue contrast, making it almost impossible to accurately localize tumors in the soft tissue, such as liver tumors.

This dissertation presents the improvements of CBCT image quality for better outcome prediction and target localization by developing the deep learning and finite element based image enhancement model.

A deep learning based CBCT image enhancement model was developed to improve the radiomic feature accuracy. The model was trained based on 4D CBCT of ten patients and tested on three patients with different tumor sizes. The results show that 4D CBCT image quality can substantially affect the accuracy of the radiomic features, and the degree of impact is feature-dependent. The deep learning model was able to enhance the anatomical details and edge information in the 4D-CBCT as well as removing other image artifacts. This enhancement of image quality resulted in reduced errors for most radiomic

features. The average reduction of radiomics errors for 3 patients are 20.0%, 31.4%, 36.7%, 50.0%, 33.6% and 11.3% for histogram, GLCM, GLRLM, GLSZM, NGTDM and Wavelet features. And the error reduction was more significant for patients with larger tumors. To further improve the results, a patient-specific based training model has been developed. The model was trained based on the augmentation dataset of a single patient and tested on the different 4D CBCT of the same patient. Compared with a group-based model, the patient-specific training model further improved the accuracy of radiomic features, especially for features with large errors in the group-based model. For example, the 3D whole-body and ROI loss-based patient-specific model reduces the errors of the first-order median feature by 83.67%, the wavelet LLL feature maximum by 91.98%, and the wavelet HLL skewness feature by 15.0% on average for the four patients tested.

In addition, a patient-specific deep learning model is proposed to generate synthetic magnetic resonance imaging (MRI) from CBCT to improve tumor localization. A key innovation is using patient-specific CBCT-MRI image pairs to train a deep learning model to generate synthetic MRI from CBCT. Specifically, patient planning CT was deformably registered to prior MRI, and then used to simulate CBCT with simulated projections and Feldkamp, Davis, and Kress reconstruction. These CBCT-MRI images were augmented using translations and rotations to generate enough patient-specific training data. A U-Net-based deep learning model was developed and trained to generate synthetic MRI from CBCT in the liver, and then tested on a different CBCT dataset.

Synthetic MRIs were quantitatively evaluated against ground-truth MRI. On average, the synthetic MRI achieved 28.01, 0.025, and 0.929 for peak signal-to-noise ratio, mean square error, and structural similarity index, respectively, outperforming CBCT images on the three patients tested. To further improve the robustness of synthetic MRI generation, we developed an organ specific biomechanical model. This model registers the pretreatment MRI images to onboard CBCT images based on the organ contours, and combines the MRI organ with CBCT body to the generate hybrid MRI/CBCT. 48 registration cases were performed, which includes 18 Monte Carlo simulated cases and 30 real patient cases. We identified tumor landmarks of hybrid MRI/CBCT, onboard CBCT and planning CT, and calculated errors of landmark locations of two CBCT images. The errors were calculated based on the landmark differences of two CBCT images and ground truth planning CT. The results show that the tumor landmark localization accuracy around tumor is improved by 54.2 ± 22.2 %.