Electronic Supplementary Material
Description and validation of the Noise Simulatortool
The Noise Simulator (Toshiba Medical Systems, Otawara, Japan), used in this study, is a tool that is based in the raw image data domain and takes into account scanner noise characteristics. This is unlike other image based noise simulators where Poisson or Gaussian noise is added to the reconstructed image. To achieve quantitative and accurate results, scanner noise characteristics are initially determined by taking multiple water phantom scans from 7.5mAs to 375mAs. These results are used to determine the amount of pixel noise that needs to be added to the acquired projections if a lower tube current is used. Once these new simulated projections are generated, this data is used to reconstruct the images that correspond to the lower tube current scans.
The "Noise Simulator" software was validated by placing a commercially available plaque phantom (QRM, Moehrendorf, Germany) within a commercially available adult thoracic anthropomorphic phantom (Lungman, Kyoto Kakagu, Kyoto, Japan), and scanned using a dynamic volume CT (Aquilion One, Toshiba Medical Systems, Otawara, Japan). The scan parameters were: 320x0.5mm detector configuration, 120kVp, 350ms/rotation gantry speed, and eight different tube current potentials: 300, 150, 100, 80, 50, 40, 30, and 20mA (mAscan). All reconstructions were performed with the same parameters using 3mm trans-axial slice thickness without reconstruction overlap and a medium-smooth Kernel (FC12).
The 300mAscan raw data acquisition was used in the “Noise Simulator” software which introduced noise in order to generate new data sets corresponding to simulated tube currents, mAsim, of 150, 100, 80, 50, 40, 30, and 20mA. Each projection dataset generated from either real or simulated acquisitions was used to generate 3mm trans-axial images. All images generated from the simulated data sets (mAsim = 150, 100, 80, 50, 40, 30, and 20mA) were compared to the images generated from real data acquisition with tube current mAscan= mAsim. In the accompanying figures, we display one of the images generated at tube currents of 150, 100, and 50mA. Images A-C correspond to real data acquisition, while images D-F are simulated data. Qualitatively, it is difficult to distinguish between the scanned and simulated images even at 50mA (Fig.C compared to Fig. F). For quantitative analysis, five regions of interest (ROI) are prescribed to allow comparison of mean CT number and standard deviation in Hounsfield Units (HU) between scanned and simulated data. These results are presented in the table as CT Number value ± Standard Deviation (HU). We have also calculated unpaired t-test p-values for each ROI to compare the simulated and scanned image results. All p-values are > 0.1 which indicate that the difference in CT numbers is not statistically significant. These results indicate that the simulated data accurately match real data acquired even at tube currents as low as 50mA (17.5mAs).
Fig Appendix
FigAppendix: Quantitative comparison of image noise between simulated and real image data using commercially available plaque and anthropomorphic chest phantoms and the “Noise Simulator” software. Images A-C correspond to images reconstructed with real data acquisition using an actual tube current, mAscan, of 150, 100 and 50mA for A, B, and C; respectively. Images D-F correspond to images reconstructed by applying the “Noise Simulator” tool to scanned raw data at 300mA and creating transaxial images corresponding to simulated tube currents, mAsim, of 150, 100, and 50mA for D, E, and F; respectively. All reconstructions were performed with the same parameters as Ca Score: 3mm slice thickness without overlap and medium-smooth Kernel (FC12). For each image, five regions of interest (ROI) are prescribed to allow comparison of mean CT number and standard deviation in Hounsfield Units (HU) between scanned and simulated data. These results are displayed in the table as CT Number value ± Standard Deviation (HU). We have also calculated unpaired t-test p-values for each ROI to compare the Simulated and Acquired results. All p-values are > 0.1 which indicate that the difference in CT numbers is not statistically significant.