Skip to main content
eScholarship
Open Access Publications from the University of California

Characterizing and Minimizing the Impacts of Diagnostic Computed Tomography Acquisition and Reconstruction Parameter Selection on Quantitative Emphysema Scoring

  • Author(s): Hoffman, John
  • Advisor(s): McNitt-Gray, Michael F
  • et al.
Abstract

Computed tomography (CT) has proven to be a critical component of clinical care, and at present there is a strong interest in quantitative imaging: augmenting or assisting human readers through the use of quantitative and computational techniques applied to the image data. While quantitative imaging is extremely promising and has been the focus of many research projects, widespread clinical adoption has not yet occurred, in part due to the susceptibility of many tests to CT acquisition and reconstruction parameters (such as radiation dose, reconstruction kernel, reconstruction algorithm, etc.). Previous efforts to illustrate and quantify the effects of parameter selection on various quantitative imaging tests have been limited in their ability to inform broader use of quantitative imaging; this is partly due to the number of parameters investigated (typically one) and/or the cohort size. This work builds on previous efforts by studying one well-established quantitative imaging test, namely emphysema scoring, using a newly-developed, high-throughput, quantitative imaging pipeline. Because of the number of conditions investigated and the cohort side, a key goal of this work is to provide recommendations for the clinical use of quantitative emphysema scoring.

The high-throughput pipeline was utilized to reconstruct a cohort of 142 subjects, scanned using the lung-screening protocol at our institution. Each scan was reconstructed under a variety of conditions: 100%, 50%, 25%, and 10% dose levels; 0.6mm, 1.0mm, and 2.0mm slice thickness; and smooth, medium, and sharp reconstruction kernels. Additionally, two reconstruction approaches were investigated: weighted filtered backprojection (wFBP), and an implementation of iterative reconstruction (Siemens SAFIRE). Thus, each scan was reconstructed using 72 unique parameter configurations.

First, the susceptibility of quantitative emphysema scoring was investigated and characterized by determining “safe” parameter configurations (i.e. resulting in small emphysema score change from a reference value computed on the 1.0mm, smooth kernel, 100% dose, wFBP reconstruction). Second, an adaptive denoising method (bilateral filtering, adjusted based on slice thickness and dose) was applied and safe parameter configurations were reassessed.

It was found that there exist small groupings of parameter combinations near the reference value that produce quantitative emphysema scores similar to the reference. This suggests that careful protocol adherence is not strictly necessary to obtain a reasonably accurate quantitative emphysema score, however there were still many parameter configurations that resulted in large deviations from the reference score. In terms of clinical translation, this suggests that in addition to a standardized, recommended protocol, one or two small changes would not typically compromise the results. However, if a scan or reconstruction was acquired using a parameter configuration deemed “unsafe,” this approach provides no means to obtain a valid emphysema score, other than to reacquire or re-reconstruct the data, which is not typically available.

With adaptive denoising applied, substantially more parameter configurations were found to result in acceptable levels of change. Only the parameter configurations using 10% dose resulted in problematic emphysema score changes. Thus, adaptive denoising provides a means to greatly improve the reliability of quantitative emphysema scoring, most importantly in cases where the scan or reconstruction fall outside of typically accepted standards.

While there is still more investigation needed, this dissertation illustrates that widespread quantitative emphysema scoring could be made more viable via the use of adaptive denoising, and in the absence of denoising only some parameter configurations yield acceptable quantitative results. Additionally, the high-throughput pipeline discussed can be applied to future, similar investigations regarding emphysema scoring, as well as investigations into other quantitative or computational imaging techniques.

Main Content
Current View