Patient-specific organ-based dosimetry and image quality estimation in clinical computed tomography

Abstract

<p>The purpose of this dissertation was to investigate and demonstrate the feasibility of patient organ based radiation dose and image quality assessment in clinical computed tomography (CT). The clinical wide usage of CT necessitates its safety and quality monitoring in a patient-centric and precise manner. This dissertation enabled this goal by developing novel image assessment frameworks and demonstrating their clinical utilities in population studies, specifically in three parts: 1) developing and clinically implementing a patient-informed radiation dose quantification framework, 2) developing a patient-specific radiation dose quantification framework, and 3) developing and clinically utilizing a patient-specific organ-based dose and image quality assessment framework.Specifically, in part one, a patient-informed organ dose estimation framework was developed and implemented for clinical imaging of adults, pediatric, and pregnant patients across a comprehensive range of scan protocols. The framework parametrically matches the patient to an anatomical phantom and quantifies the radiation field using a regional volume CT dose index (CTDIvol) based approach. The estimation further incorporates an accuracy model to calibrate bias and quantify uncertainties. The framework was clinically implemented to investigate the differences between organ dose based effective dose with dose length product based effective dose, which is non-patient, scanner output derived, for 1048 patients. The results show the organ-based measurements are significantly different from the non-patient based, especially for patients with relatively small or large sizes. In part two, a patient-specific approach, in contrast to the patient-informed approach, in organ dose assessment clinical CT was explored. A novel framework, iPhantom, was developed to automatically generate patient-specific phantoms (i.e., anatomical digital-twins) based on CT images. Specifically, anchor organs with relatively distinguishable contrast (i.e., lungs, bones) are segmented using a deep learning-based multi-organ segmentation model. Other radiosensitive organs that are challenging to segment (i.e., intestines) are filled in from an anatomical template using affine and diffeomorphic registration. The framework is combined with a Monte Carlo CT simulation program and systematically validated for its utility to patient-specific organ dose estimation. The results demonstrate that the framework, for the first time, enables sufficient and automated patient-specific organ dose assessment in clinical CT. In part three, a comprehensive clinical image assessment framework was developed integrating the patient-specific organ dosimetry and in vivo image quality quantification. The framework was used to demonstrate the necessities of organ-based measurements in contrast to their commonly used non-organ-based counterparts; and to investigate the relationships of dose, image quality, and patient sizes for exams scanned with different parameters among 9801 clinical CT images. The results show that, in general, compared to the organ based, non-organ-based dose and quality metrics significantly misrepresent patient-specific values. The relationships between the two types of quantities in patient dose, image quality, and size are weak and vary substantially for different scan parameters. This study implies that although it is commonly assumed that increased dose results in superior image quality, image quality and dose at the level of patient may need to be optimized semi-independently from one another. In conclusion, this dissertation demonstrated the necessity and feasibility of automated and patient-specific safety and quality assessment in clinical CT, paving the way for patient-centric monitoring and large-scale analysis for personalized precision medicine. </p>