This work is relevant to investigators who intend to use Apparent Diffusion Coefficient (ADC) measurements as a quantitative biomarker for tumour response to drug therapy, where study design includes multiple sites, vendors and MRI field strengths. The QuIC-ConCePT¹ project is a pan European collaboration. ADC is the decay constant for signal loss between diffusion-weighted MRI images (DWI) of increasing diffusion sensitivity (image b-value). Lower ADC indicate fluid restriction within a cell-dense tumour. Following successful treatment, a tumour undergoes cell death and necrosis. Significant increase in ADC is a potential imaging biomarker for early treatment response in drug trials. Large trials recruit from multiple sites, where scanners and field strengths may vary. It is important to establish a standardised DWI acquisition and ADC analysis method that is consistent and reproducible. In our previous work using 1.5 T only², we compared different ROI methodologies using freely available software (Osirix) and developed a statistical model to estimate expected errors for individual measurements. We concluded that ROI volume inversely affects reproducibility regardless of methodology. As part of QuIC-ConCePT, a software platform has been developed (Keosys Imagys) that allows all collaborators to upload data for central analysis. Our ADC analysis methods³ have been incorporated into a standardised region definition and analysis tool that can directly compare multi-site ADC reproducibility of 3D regions at 1.5 T and 3 T.Following ethical approval and informed consent, data was acquired from the 5 sites (vendor details and short hand in parenthesis); VU University Medical Centre, Amsterdam (Site A, GE Signa HDxt 1.5 T), University of Manchester (Site B, Philips Achieva 1.5 T), Radboud University Nijmegen Medical Centre (Site C, Siemens Magnetom Avanto 1.5 T and Siemens Trio 3 T), Institute of Cancer Research, Royal Marsden Hospital, London (Site D, Siemens Magnetom Avanto 1.5 T), French Institute of Health and Medical Research, INSERM, Paris (Site F1,Philips Ingenia 3.0T and Site F2, GE Discovey 3 T). Patients were scanned twice within 14 days. Manual whole tumour ROIs were defined on b-100 (1.5 T) or b-150 (3 T) images, (excluding first and last slices to minimise partial voluming). A single lesion was chosen based on size and location (right lobe, away from the heart or diaphragm where possible). Coefficent of Variance (CoV) of test-retest volumes was used to compare the software platform to previous Osirix-based analysis. Voxel ADC values were calculated from the mono-exponential fit (corrected for high b-value SNR bias) from 3 b-value images (100, 500, 900 s/mm2 for 1.5 T and 150, 400, 800 s/mm2). ADC metrics (mean, median, ΔADC%) were calculated from the ROI histograms. Reproducibility was measured using CoV. Average fit failure rates between sites and field strengths were compared. Datasets were categorised according to quality. Lesions subjected to visible motion, low SNR, poorly positioned (sub-phrenic or right heart border) or cystic, were identified prior to analysis. Bland-Altman style plots of tumour volume against percentage change in ADC (ΔADC%) were compared between field strengths.32 patients were recruited between 2012 to 2015 (M:F ratio 4.33:1, median age 64, range 44-77). Recruitment per site (1.5 T) was; Site A (5), Site B (5), Site C (5), Site D (5). Recruitment per site (3 T) was; Site C (4), Site F1 (5), Site F2 (3). The Imagys platform compared favourably to previous work where tumours were defined on Osirix (CoV 8% vs 9%). At 1.5 T and 3 T respectively, average absolute mean ADC was 112 x 10-5 mm2/s (range 79 – 214 x 10-5 mm2/s) and 140 x 10-5 mm2/s (range 85 – 214 x 10-5 mm2/s). Overall reproducibility at 3 T also compared favourably to 1.5 T (CoV 6.4% vs 7.3% for mean ADC). The percentage of voxel ADC fit failures was significantly higher for 3 T data. Bland Altman style plots show a clear trend towards improved reproducibility with increasing tumour volume (i.e. sample size). Datasets subjectively categorised according to quality are highlighted.We expect reproducibility to be dependent on tumour size, SNR, tissue heterogeneity and motion¹. Overall reproducibility of ADC (mean, median) was similar between field strengths. Site F2 (n=3) enhances overall 3 T CoV, as two data sets were mostly cystic (homogenous). Lesions with very low SNR have a similar profile, i.e. appear homogenous. At both field strengths, the majority of outliers were subject to motion. The furthest outlier (3 T) had a fitting error of 30% despite no obvious quality issue. 3 T data is affected by ADC fit failure. Further work is required to improve the fitting routine for 3 T datasets.