Scholars@Duke publication: Robustness and Accuracy of An Automated Solution for Measuring Image Noise, Spatial Resolution, Contrast, and Dose in Clinical CT Images: Efforts Towards Patient-Specific Quantifications of CT Image Quality and Radiation Dose

Robustness and Accuracy of An Automated Solution for Measuring Image Noise, Spatial Resolution, Contrast, and Dose in Clinical CT Images: Efforts Towards Patient-Specific Quantifications of CT Image Quality and Radiation Dose

Publication , Conference

Ding, A; Smith, T; Abadi, E; Ria, F; Wilson, J; Samei, E

Published in: Medical Physics

June 30, 2017

Purpose: To develop a robust and accurate automated solution for measuring image noise, spatial resolution, contrast, and radiation dose using clinical patient CT images. Methods: An automated infrastructure has been developed to measure the image quality and radiation dose metrics from daily clinical CT images data routed through the picture archiving and communication system (PACS). Previously-validated algorithms for measuring image noise, CT resolution index, and organ-based Hounsfield units (HUs) were re-compiled and integrated with the infrastructure. All scan- and dose-related entries are automatically extracted from the DICOM header metadata. Multiple regions of interest (ROIs) are segmented and HUs are characterized. The noise values are calculated by sampling the peak histogram value of the soft tissue isolated ROIs. The CT resolution index (RI) values are extracted to calculate the edge-spread function (ESF). Using an IRB-approved protocol, the robustness of this metrology was evaluated by counting the successful outputs across a total of 1000 patient cases from contrast and non-contrast enhanced CT chest/abdomen/pelvis exams with different reconstruction algorithms and reconstruction kernels. The accuracy was assessed by comparing the results with previously published data. Results: The automated infrastructure is successfully developed to measure the image quality and dose metrics using patient CT data. All the dose (CTDIvol and DLP) and noise data were successfully calculated. HUs measurements and noise assessment algorithms exhibited over 95% success rate, which RI algorithm yielded a 99% success rate. An overall accuracy of 90% was detected in the results. Conclusion: Patient-specific image quality and dose metric can be quantified by using an automated solution from clinical datasets. The method can be applied to daily CT image quality and dose monitoring to improve overall diagnostic performance.