Ranking of Mobile C-Arm Image Quality Using Contrast–Detail Detectability Metrics
Abstract
Purpose
In this study, image quality of eleven mobile C-arm fluoroscopic systems across one institution was evaluated using visual assessment and Rose Signal-to-Noise Ratio (SNR) calculation. The aim was to rank c-arm image quality using contrast detail-based metrics.
Methods
A contrast–detail phantom (Sun Nuclear Model 1151) placed onto a 2.4mm Cu sheet was imaged using fixed geometry for all systems. Displayed Air Kerma Rate was recorded. Visual assessment was performed in a radiology reading room using a medical-grade display using the default window width and level settings. Disk visibility was scored by (a) two observers viewing the entire phantom image and (b) one observer viewing individual disks by use of a mask covering most of the phantom, with five repeats. Rose SNR was calculated for each disk (SNR=(μs−μb)/(σb )*√Ad). For all three methods, total number of detected targets was determined and contrast-detail curves were constructed. Threshold contrasts were fit to a model function, f(r) = k/r2+b where r denotes radius; k, proportionality; b, bias; f, threshold contrast. The correlation of the rankings was computed using Kendall’s tau.
Results
The image quality of the flat-panel based C-arm was clearly superior as threshold contrasts were lowest at all disk diameters. The AKR of all image-intensifier-based C-arms (10 out of 11) was similar. Contrast-detail curves followed similar trends but varied on an absolute scale. Scoring while viewing the entire phantom resulted in lower threshold contrasts than the mask-based scoring. The total number of detected targets produced moderately consistent rankings for both human reading paradigms (τ=0.54, p=0.047). Rose-SNR did not produce consistent rankings.
Conclusion
Contrast-detail detectability is commonly used to assess radiographic image quality. Our study highlights the importance of controlling human observation conditions to reduce variability and the need for developing a numerical model to minimize subjectivity in human assessments.