Modeling and Optimization of Individualized Liver Contrast-enhanced CT Imaging

Abstract

More than half of clinical CT imaging in the United States involves the use of iodinated contrast materials. The use of such contrast agents in CT imaging enhances tissue contrast, particularly in soft tissue organs such as the liver, pancreas, spleen, and kidneys, and thus improves the depiction of a variety of disorders. Despite the critical role of contrast media administration in clinical practice, there is a lack of standardization in contrast administration techniques across institutions. As a result, many studies have indicated inconsistencies in contrast enhancement across different patients, posing clinical diagnostic risk in over- and under-enhanced patient cases. In addition, contrast agents have been known to potentially increase the risk of contrast-induced acute kidney injury (CI-AKI) and cause allergic reaction, such as urticaria or anaphylactic shock in small number of patients. Therefore, given the diagnostic benefit, risks, and the prevalence usage of iodinated contrast agent in CT imaging, there is a need to devise a mechanism to increase the consistency and adequacy of organ contrast enhancement through the personalization of contrast administration and scanning parameters for each patient according to their known pre-scan attributes.This dissertation primarily focuses on liver imaging as clinically, liver is one of the most vulnerable and susceptible organs to diffuse diseases (fatty liver, hepatitis, cirrhosis), benign tumors, cancers (especially hepatocellular carcinoma), and metastasis development from non-liver primary cancers. While there are other modalities which can be used to image liver, contrast-enhanced CT imaging is the most commonly used technique to screen for these abnormalities in both healthy and ill individuals as improvement of the disease outcome relies on the accuracy of early diagnosis from the screening. The purpose of this dissertation was to demonstrate the feasibility and clinical utility of building a patient-informed hepatic parenchyma contrast enhancement prediction model using retrospective clinical images. This project was conducted in two parts: 1) constructing the patient-informed contrast enhancement prediction model using retrospective patient cases and prior knowledge, 2) implementing a prospective preliminary clinical test using the prediction model on limited number of patients. The first part of the dissertation (chapters 2 and 3) covers both the preparatory works in building the contrast enhancement prediction model and the process of building the model itself. In chapter 2, we used a small library of patients to explore the feasibility of potentially building a liver parenchyma contrast enhancement model. Some of these works included determining any correlations between the available patient attributes (height, weight and wight-derivative factors, sex, age) and the contrast enhancement HU level of the patient at the time of scan (portal venous phase). The correlation we observed enabled us to build a machine learning model using patient attributes to predict the contrast enhancement level of the liver at the time of scan. In chapter 3, we expanded our patient library and introduced a method of building a more robust contrast enhancement model using Gaussian function and neural network. In the second part of the dissertation (chapter 4), we turned our focus to applying the patient-informed prediction model prospectively in the clinic. Chapter 4 involved the development of a graphical user interface which contains the prediction model of the previous chapter to provide prospective, real-time prediction of a patient’s hepatic parenchyma contrast enhancement level given patient attributes and the starting CT number value of the parenchyma. To make the tool useful in the clinical setting, the prediction model was paired with an optimization algorithm to determine alternative injection and scanning protocol in real time, targeting changes needed to the scanning parameters of patients with predicted under- or over-enhancement of the hepatic parenchyma at the time of scan. A pilot study of 24 patients undergoing contrast-enhanced CT imaging with Renal Cell Carcinoma (RCC) scanning protocol was conducted to assess the feasibility of using the aforementioned patient-informed tool to inform scanning and contrast injection parameters of each patient. For 19 of the 24 patients, we also compared the outcome of this study (contrast enhancement level) with their previous conventional RCC- protocol or abdominal contrast-enhanced CT imaging. We concluded the dissertation with summative conclusions, clinical implications, and potential future directions of this project. In conclusion, this dissertation developed a patient-informed liver parenchyma prediction model using retrospective clinical cases and explored the potential benefit of implementing such model in the clinic through a feasibility clinical study.