A Novel Methodology for Automated Image Quality Assessment of Pediatric CT
PURPOSE To develop an effective and automated methodology for patient-specific image quality assessment of clinical pediatric body CT examinations. METHOD AND MATERIALS This IRB approved study evaluated 816 clinically performed (6/14-11/17), contrast-enhanced abdominopelvic (AP) CT scans of pts ages 0-18. Studies were from 3 scanners: Siemens Flash (n=637), GE 750 HD (n=72), and GE VCT (n=107). Quality metrics included noise magnitude, spatial resolution, and image contrast enhancement. Previously validated methods to measure these in adults were modified and validated for use in pediatric clinical CT images. For noise, the algorithm uses uniform areas of the images to characterize the standard deviation (SD) of Hounsfield units (HU). For contrast, the algorithm identifies ROIs in the liver, lung, aorta, and spine and measures HU values and SD. For resolution, the algorithm samples the skin-air interface which allows for quantification of spatial resolution. The frequency at 50% of the MTF curve is used in our resolution reporting which was found concordant with observer performance. RESULTS This system allows for determination of quality variability between any divisions of patient- or scanner-specific metrics (e.g., age, size, gender, model, protocol). In one scanner, between the youngest and oldest age groups, noise increased by 89%, absolute contrast in the liver increased by 14%, the standard deviation of liver contrast increased by 129%, and the f50 value for resolution decreased by 6% (Figure). CONCLUSION A novel automated system to obtain advanced quantitative image quality metrics in clinically performed AP examinations in children was established. We are able to quantitatively assess differences between metrics of study quality from patient studies rather than using simplistic phantom studies, observer studies, or manual determination of quality. One benefit of this model is that these patient-specific metrics can be integrated to generate a task-specific quantification to predict observer performance on lesion detection. This quantitative tool serves as a foundation for true CT performance optimizations beyond what can be achieved with the current quality assessment methods and dose monitoring alone. CLINICAL RELEVANCE/APPLICATION Current CT practice quality optimization utilizes phantom studies or cumbersome methods that require observers or manual analysis. Automated quality metrics would improve assessment of CT performance.
