Feature-guided deep learning reduces signal loss and increases lesion CNR in diffusion-weighted imaging of the liver

Multi-center and multi-vendor evaluation study across 1.5 T and 3 T scanners (part 1): apparent diffusion coefficient standardization in a diffusion MRI phantom

OBJECTIVE

To validate multi-site and multi-vendor ADC measurements using the QIBA/NIST diffusion MRI phantom at room temperature.

MATERIALS AND METHODS

ADC measurements were performed on 12 scanners (evenly split between 1.5 and 3 T) from three vendors at five sites and compared with reference values at room temperature. We adopted Pearson's correlation (r) and accuracy error for comparison with reference values; within scanner coefficient of variation (CVintra%) for intra-session repeatability and inter-scanner for agreement (CVinter%); Bland-Altman plots and precision error for short-term reproducibility; generalized linear mixed models and post-hoc tests ( α =0.05) to compare accuracy, repeatability and precision across field strengths, vendors, and scanners.

RESULTS

Temperature adjusted ADCs were well correlated with NIST reference values (r ≥ 0.997 for 1.5 T, r ≥ 0.996 for 3 T). Median accuracy error was lower than 5% for all scanners. In the renal physiologic range (ADC > 0.83 × 10-3 mm2/s), accuracy error was < 10% and CVintra < 2%. Across all scanners, good short-term reproducibility with limits of agreement < 10% and excellent agreement (median CVinter < 2%) were found.

DISCUSSION

Despite using abdominal receive coils and room temperature measurements, all quantitative parameters were within literature findings. High accuracy, repeatability and precision within the renal physiologic range support the feasibility of scanner evaluation using QIBA standardization process for diffusion measurements in renal studies.