(Sunday, 7/29/2018) 3:30 PM - 4:00 PM
Room: Exhibit Hall | Forum 8
Purpose: Quantitative imaging in CT can potentially assist in disease characterization, diagnosis, tracking, and prediction. Optimally, any variation in the measured signal is isolated to changes in the disease of interest. In practice, additional variation is introduced from the acquisition and reconstruction of the CT image itself. Changes in patient dose, slice thickness, and reconstruction kernel can add different levels of noise characteristics in the image, which has been shown to affect downstream quantitative measures [cite previous work]. We hypothesize that variation of quantitative measures as a result of different acquisition and reconstruction parameters can be mitigated via bilateral denoising techniques.
Methods: 17 nodules from independent patients, imaged under lung cancer screening protocols, were analyzed. Each scan had raw projection data collected from the scanner, and reconstructed at different conditions (relative doses=10%, 25%, 50%, 100%; slice thickness=1.0mm, and medium reconstruction kernel) using an in-house reconstruction pipeline, which employs a previously validated noise-addition model to simulate low-dose acquisition. Reconstructions were done using traditional weighted filtered back projection and iterative reconstruction methods (WFBP and SAFIRE, respectively). For comparison of the effects of denoising on quantitative metrics, these datasets were also denoised using bilateral filtering (WFBPd and SAFIREd, respectively). An in-house CAD system detected and segmented nodules, which were confirm to the aforementioned radiologist-approved detections. Features – GLCM entropy, standard deviation, and mean – were extracted from CAD segmentations of each nodule in each data subset (WFBP, WFBPd, SAFIRE, SAFIREd).
Results: The standard deviation feature, with varying distributions across dose levels, is improved by denoising in WFBP and has similar absolute differences in SAFIRE. Features with similar distributions across doses (GLCM Entropy, mean HU) have minimal negative effects from denoising.
Conclusion: Denoising can assist in bringing quantitative values towards similar distributions, which is demonstrated in this study across varying dose levels.
Funding Support, Disclosures, and Conflict of Interest: Funding support for this research was provided in part by the University of California Office of the President Tobacco-Related Disease Research Program (UCOP-TRDRP grant #22RT-0131) and the National Cancer Institute's Quantitative Imaging Network (QIN grant U01-CA181156).
Keywords
Quantitative Imaging, Texture Analysis
Taxonomy
IM- CT: Quantitative imaging/analysis
Contact Email
