Fewer Samples, Equally Confident: An Adaptive Sampling Method for Estimating the Coefficient of Variation with Sample Size Based QC Limits
Presentations
PO-GePV-I-91 (Sunday, 7/10/2022) [Eastern Time (GMT-4)]
ePoster Forums
Purpose: Exposure reproducibility of radiography equipment is typically quantified by the coefficient of variation (CV). Federal regulations stipulate that the CV be estimated as the ratio of sample standard deviation to sample mean using 10 samples. However, the number of samples (n) varies in practice, for example, Michigan specifies 4 while Texas has no guidance, but both share the federal limit of CV ≤ 0.05. Practically, CV is a biased estimator, and its confidence limits depends heavily on n. This abstract quantifies the effect of n on CV, suggests federal equivalent limits for n < 10, and demonstrates the use of these equivalent limits on real QC data to facilitate adaptive sampling.
Methods: Ninety-five percent confidence limits were generated for CV from 0 to 0.05 using Vangel modified McKay method from n = 3 to 10. For n < 10, CV values with the same upper confidence limit as n = 10 were determined as federal equivalent limits. To demonstrate the effect of these equivalent limits, 111 exposure reproducibility measurements, each with n = 10, were resampled with replacement 10,000 times and averaged to calculate their CV for n from 3 to 9.
Results: Federal equivalent limits were 0.015, 0.024, 0.032, 0.037, 0.041, 0.045, and 0.048 from n = 3 to 9. Of the 111 reproducibility measurements, 106 require only 3 samples to pass, while 2 require 4 and the rests require 4, 5, 6, and 8 samples each, representing a 69% reduction in the sample size required for the same confidence in reproducibility when adaptive sample size technique was used.
Conclusion: Exposure reproducibility compliance can be determined using fewer samples, but care needs to be taken to adjust the passing criteria. This can lead to more efficient QC by using fewer samples and only acquiring additional samples as needed.
Keywords
Taxonomy
IM- X-Ray: Quality Control and Image Quality Assessment
Contact Email