TH-E-217BCD-09: Task-Based Image Quality of CT Iterative Reconstruction Across Three Commercial Implementations

Abstract

Purpose: Iterative reconstruction (IR) has the potential to reduce patient dose while maintaining image quality comparable to filtered back projection (FBP). There are several different IR algorithms, however, with each offering different resolution and noise texture. The goal of this project is to compare image quality of FBP and IR images across three CT scanner models using metrics that incorporate system resolution and noise texture. Methods: The American College of Radiology CT quality assurance phantom was scanned using three CT scanner models (model A: GE Discovery CT750 HD, model B: Philips iCT, and model C: Siemens FLASH). Images acquired at seven dose levels ranging from 1 to 20 mGy were reconstructed using both FBP and IR. The data acquisition was randomized and duplicated five times to reduce the effect of systematic variables. The MTF was measured from the edges of the cylindrical inserts in the phantom. Square regions taken from the uniform section of the phantom were used to compute the NPS. From the MTF and NPS for each series, the d' was computed for a 5 mm object. Results: At a typical clinical dose level of 12 mGy, the difference in MTF10 between IR and FBP was 5%, 5%, and 8% for models A, B, and C, respectively. Additionally, the peak of the NPS for IR, when compared to FBP, was shifted to lower frequencies by 30%, 13%, and 22% for models A, B, and C respectively. Preliminary detectability index values were 20–40% higher for IR compared to FBP. Conclusions: Using IR increased the d' compared to FBP for all scanner models. The percent dose reduction achievable while maintaining image quality, however, varied across scanner models. This data provides a new perspective to define and optimize CT protocols across commercial scanner models.