Room: Davidson Ballroom B
Purpose: Longitudinal assessment of quantitative imaging biomarkers (QIBs) requires a comprehensive quality control (QC) program to minimize bias and variance in the measurement results. In addition, the availability of data analysis software from multiple vendors suggests the need for a means of quantitatively comparing the QIB outputs of the applications. The purpose of this work is to describe a digital reference object (DRO) that has been developed for the evaluation of arterial spin-labeling (ASL) measurement results.
Methods: The ASL DRO is a synthetic data set consisting of 10x10 voxel square blocks with a range of SNR and blood flow (BF) values (SNR:1-100, BF:10-210 ml/100g/min). Matlab code was developed to simulate a pseudo-continuous ASL sequence with parameters and signal intensities defined according to clinically-acquired images. Rician noise was added to the DRO to achieve a range of SNR values based on the ASL control image data. ASL parameters were estimated using the commercially-available nordicICE software package (NordicNeuroLab, Inc, Milwaukee, WI). Calculated parameters were compared against ground truth with percent bias measures, Bland-Altman analyses, and concordance correlation coefficients (CCCs).
Results: Measurement bias and variability improved with increasing SNR and BF values. Across all SNR and BF values, the average bias was 8.1%Â±14.3 with a median COV of 4.8%. Considering limited SNR and BF values (SNR:10-100, BF:40-210 ml/100g/min), the average bias was 5.1%Â±1.0 with a median COV of 3.5%. Excellent agreement with reference values was seen for all BF values above an SNR of 10 (CCC greater than 0.93).
Conclusion: The ASL DRO developed in this work allows for the evaluation of software bias and variance across physiologically-meaningful BF and SNR values. Future work will involve evaluating the range of SNR and BF levels observed in clinical ASL images to establish the relevant DRO parameter space to be used in ongoing QC programs.