Abstract
BACKGROUND: Diffusion-weighted imaging (DWI), a quantitative magnetic resonance imaging (qMRI) technique, has the potential to aid in disease characterization and treatment response monitoring. MR-Linacs (MRLs) enable simultaneous DWI acquisitions during radiotherapy, uniquely aiding in the collection of large-scale datasets for imaging biomarkers, such as the DWI-derived apparent diffusion coefficient (ADC), without additional patient burden. However, the limited data reporting on variability in MRL scanner performance characteristics, and a lack of established clinical trial quality assurance (QA) procedures, are barriers to this route for biomarker validation. PURPOSE: This study aims to quantify the accuracy, intra-scanner repeatability, and inter-scanner reproducibility of ADC measurements across three MRLs in Australia in both a phantom and in vivo. These measurements will inform the feasibility of carrying out prospective multi-center studies in Australia investigating ADC as a biomarker and form a core set of QA procedures and baselines to assess biomarker and sequence suitability. METHODS: An isotropic diffusion phantom (at 0°C) and one healthy volunteer were scanned on three Unity MRLs (Elekta AB, Stockholm, Sweden). Standardized (QIBA Diffusion Profile) and anatomy-specific DWI sequences, including sequences recommended by the MR-Linac Consortium Imaging Biomarker Working Group, were used to image the phantom and volunteer. ADC maps generated using the MRL scanner software (inline ADC) and diffusion-weighted (b-value) images were exported from the scanner console. The latter was used to generate ADC maps using commercial software (offline ADC) for a separate comparative analysis. Performance metrics were computed for each sequence, including a coefficient of variation to assess between-session intra-scanner repeatability (CV(BS)) and inter-scanner reproducibility (CV), for each phantom vial and contoured organ. Additionally, using the phantoms' known ADC vial values, a percentage bias (bias) was calculated to determine ADC accuracy. RESULTS: Phantom-based measurements for the standardized QIBA sequence had intra- and inter-scanner CV and bias well within recommended guideline (QIBA Diffusion Profile) tolerance limits of 2.2% and ±3.6%, respectively. All anatomy-specific phantom DWI sequences were also within these tolerances, except for the cervix sequence at one site which showed an average intra-scanner bias of +4.5%. Both accuracy and reproducibility for all sequences were worse for lower diffusivity vials measured in the phantom. Additionally, inline and offline ADC maps had high similarity with average percent differences of +0.2%. Volunteer-based results had worse reproducibility, with the average inter-scanner CV for the brain and pancreas sequences within 9.0%, however, reaching up to 27.1% for pelvis and abdomen sequences. CONCLUSIONS: This study demonstrated accuracy, intra-scanner repeatability, and inter-scanner reproducibility comparable to metrics reported in the literature, using both the phantom and volunteer datasets. The cervix sequence had the largest variability in both phantom and volunteer results and was recommended for further investigation. This study suggests that qMRI techniques utilizing DWI could be a viable option for future multi-centered patient-based studies utilizing Australian MRLs, with phantom-based quality assurance recommended alongside patient imaging.