Comparison of visual and semi-automated kilovoltage cone beam CT image QA analysis.

Established kilovoltage cone-beam computed tomography (kV-CBCT) image quality assurance (QA) guidelines often rely on recommendations provided by the American Association of Physicists in Medicine (AAPM) task group (TG) reports with metrics that use visual analysis. This can result in measurement variations by different users, especially in visually subjective analyzes such as low contrast resolution. Consequently, there is a growing interest in more automated means of image QA analysis that can offer increased consistency, accuracy, and convenience. This work compares visual QA to semi-automated software QA analysis to establish the performance and viability of a semi-automated method. In this study, a commercial product (RIT Radia. Radiological Imaging Technology, Colorado Springs, CO) was used to evaluate 68 months of kV-CBCT images of a Catphan® 504 phantom obtained from a Varian TrueBeam® linear accelerator. Six key metrics were examined: high contrast resolution, low contrast resolution, Hounsfield unit constancy, uniformity and noise, and spatial linearity. The results of this method were then compared to those recorded visually using Bland-Altman, and/or paired sample t-test. Comparison of all modules showed a non-random, statistically significant difference between visual and semi-automated methods except for LDPE and Teflon in the Hounsfield unit constancy analysis, which falls outside the paired sample t-test's 5% significance level. A small high contrast resolution bias indicates the two analysis methods are largely equivalent, while a large low contrast resolution bias indicates greater semi-automated target detection. Wide limits of agreement in the uniformity module suggests variability due to multiple visual observers. Spatial linearity results measured differences of less than 0.17%. Semi-automated QA analysis offered greater stability over visual analysis. Additionally, semi-automated QA results satisfied or exceeded visual QA passing criteria and allowed for fast and consistent image quality analysis.