Radiation Dose-Dependent Characteristics of CT Images Reconstructed with Statistical Iterative Algorithms
L Genovese*, R Price, D Pickens, Vanderbilt Medical Center, Nashville, TNSU-E-I-53 Sunday 3:00PM - 6:00PM Room: Exhibit Hall
Purpose: To characterize the dependence of image noise, contrast-to-noise and noise structure in CT images on algorithm weighting-factors (AWF) and radiation dose for three different commercially available iterative reconstruction algorithms.
Methods: Images were acquired using the 20 cm ACR CT accreditation phantom on scanners manufactured by three different vendors at four different CTDI levels from approximately 10 mGy-70 mGy. Images were reconstructed using conventional filtered backprojection (FBP) image reconstruction and with each vendor's proprietary statistical iterative reconstruction (SIR) algorithms: ASiR™, SAFIRE™ and iDOSE™. At each CTDI level, images were reconstructed with multiple AWF. Image noise (standard deviation of background near contrast objects), contrast-to-noise (CNR) and Noise Power Spectrum (NPS) were plotted as a function of both weighting factor and CTDI.
Results: All SIR algorithms demonstrated significant improvement in both noise reduction and CNR, typically 30%-60% depending upon the AWF. NPS was found to be a function of both reconstruction kernel and SIR weighting factor. All SIR reconstructed images showed the same general shift in noise power to lower spatial frequencies with increased AWF regardless of vendor. With dose reduction as high as 80%, heavily weighted SIR reconstructed images can produce CNR values comparable to images produced with 100% dose levels.
Conclusion: All three SIR algorithms provide significant noise reduction and CNR improvement with minimal reduction of spatial resolution. The noise-reduction performance with different algorithm weighting was approximately linear for all SIR algorithms. For all CTDI levels measured, each algorithm was capable of providing increased CNR by at least a factor of 90%. However, the weighting factors are vendor specific, thus a direct comparison is not appropriate. The use of greater SIR weighting factors also resulted in a shift in the NPS to lower spatial frequencies, producing a background texture that could potentially affect image interpretation.