Investigation of Patient Factors That Predict Rankings of Defect Detection Performance in Pediatric Renal ^99mTc-DMSA SPECT

Objectives: The ultimate objective of this work is to provide optimal patient specific administered activities (AAs) for pediatric patients, with the initial target being DMSA renal SPECT. An optimal dose is the lowest one that provides sufficient diagnostic image quality for a given patient; in this context, image quality means performance on a renal defect detection task. Image quality is affected by several patient factors including weight, height, body morphometry, and kidney shape and depth. The goal of this study was to determine the patient factors that predict performance.

Methods: We used a mathematical observer to estimate defect detection performance for an ensemble of simulated images from a population of phantoms. We then looked at the rankings of defect detection performance and looked for factors that could be estimated from images that predict these rankings. We simulated a phantom population based on the University of Florida NHANES phantom series that included variations in age (1-15 years), gender, body habitus (50^th weight percentile and 10, 50, and 90^th height percentile), organ size (reference and +/- 15% volume), and organ uptake (from patient data). Focal acute pyelonephritis was simulated at three locations (upper pole, lateral, and lower pole) of the renal cortex. The defect volumes were chosen to be near limits of clinical detectability and to have equal contrast for the reference phantom (50^th weight and height percentile, reference kidney size) for each age. Simulated projections were scaled to a clinical count level appropriate for the current weight-based dosing protocols, reconstructed using filtered backprojection (FBP), and filtered with an order 8 3D Butterworth filter with cutoffs ranging from 0.1 to 1.0 cycles per cm. We generated 384 uptake realizations of noisy projection data for each phantom, resulting in 13,824 pairs of defect-present and defect-absent images. Channelized linear discriminant observer (CLDO) methodology was used in combination with a receiver-operating-characteristic (ROC) analysis to evaluate image quality as a function of age and height. The CLDO was applied to extracted coronal, sagittal, and transaxial images containing the defect centroid. We used 7 rotationally-symmetric, non-overlapping difference-of-mesa frequency channels with widths increasing by a power of two to model the human visual system. Internal noise was added to the test statistics to model intra-observer variability. The test statistics were analyzed using the LABROC4 code that performs maximum-likelihood ROC curve fitting. Bootstrapping and nonparametric analysis were used to compute 95% confidence intervals. We examined the variables that best predicted the performance rankings for the various phantoms.

Results: The results showed that weight alone predicted image quality rankings for a fixed patient height. However, when including phantoms with mixed heights, neither weight nor height, combined or alone, predicted image quality rankings. Instead, other factors such as anterior to posterior distance and patient girth were required.

Conclusion: This study demonstrated that, in agreement with previous results, weight alone is not a sufficient parameter for optimally scaling pediatric doses. Further, results demonstrated that the addition of height is also not sufficient for optimal dosing. Instead, other factors such as patient girth and kidney anterior and posterior depth should be considered. Research Support: This work was supported by NIBIB under grant number R01-EB013558