Photon-Counting Micro-CT: Algorithm Development and Application in Mouse Models of Disease

Abstract

Over the past fifty years, x-ray computed tomography (CT) has become a widely used imaging modality in medical applications due to its fast scanning speed and high spatial resolution. However, the contrast and energy resolution of CT has been limited by the use of energy-integrating detectors (EIDs). The recent emergence of photon-counting detectors (PCDs) for x-ray CT imaging has enabled improved image contrast, further improvements in spatial resolution, reduction of artifacts such as beam hardening, and simultaneous, radiation dose-efficient multi-energy imaging with a single x-ray source setting and detector. Importantly, the inherent multi-energy imaging capability of photon-counting CT (PCCT) supports improved material separation for current diagnostic applications as well as novel, multi contrast agent imaging procedures. Unfortunately, PCCT also comes with unique limitations. These include noisy images in high energy thresholds that necessitate the development of accurate and computationally efficient multi-energy denoising algorithms and spectral distortion effects that introduce crucial errors in the detected x-ray photon energy distribution if left uncorrected.This dissertation covers a multitude of preclinical research studies that contribute to the emerging field of PCCT by addressing its limitations and demonstrating its benefits in several disease applications. Our lab’s hybrid in vivo and ex vivo micro-CT scanners that are equipped with both PCDs and EIDs were used to acquire high-resolution images for these studies. With scans acquired on our hybrid in vivo micro-CT system as training data, we developed deep learning solutions for multi-energy PCCT denoising and spectral distortion compensation in PCCT material decomposition. We developed and applied ex vivo and in vivo PCCT imaging pipelines for quantitative studies in transgenic mice that model human risk for Alzheimer’s, cardiovascular, and bone diseases. In addition, we developed a virtual preclinical imaging trial framework to accompany PCCT imaging with multiple contrast-agents in mouse models of cancer, a technique that is used to investigate new therapies. We demonstrate that our deep learning models augmented with a novel image preprocessing strategy can provide computationally efficient multi-energy denoising while remaining adaptable to changes in PCD energy thresholds. Our deep learning model trained on data from matched PCD and triple-EID scans of ex vivo samples demonstrated the ability to produce distortion-free material maps while using low-dose, multi-energy PCCT images as its input. Our studies in transgenic mice demonstrated the unique benefits of the PCD in brain, bone, and heart imaging and revealed how combinations of genetic and lifestyle factors influence risk of disease in each of these organs. By developing a computational mouse cancer phantom and a PCCT simulation pipeline calibrated to real imaging, we show how a virtual trial framework can help optimize imaging for accurate assessment of cancer treatment efficacy. Together, these contributions show that PCCT, combined with novel computational solutions such as deep learning algorithms and virtual imaging, supports studies in a wide range of diseases due to its excellent image quality and multi-energy imaging capabilities.