Reproducibility of whole body ADC in a non-optimized multi-centre trial: Effectiveness of normalisation method
Jagadish Kalasthry1, Stuart Taylor1, David Atkinson2, Alan Bainbridge3, Shonit Punwani1, Anna Barnes4, and On behalf of STREAMLINE investigators5

1Imaging, University College Hospital, London, United Kingdom, 2Centre for Medical Imaging, University College London, London, United Kingdom, 3Medical Physics and Biomedical Engineering, University College Hospital, London, United Kingdom, 4Clinical Physics, Institute of Nuclear Medicine, University College Hospital, London, United Kingdom, 5NIHR, National Institute for Health Research, London, United Kingdom

Synopsis

Diffusion-Weighted (DW) MRI can be used as a quantitative tool but there is limited data documenting accuracy and reproducibility on different MRI scanners with non-identical protocols. We investigated the variation in ADC measurements in normal tissue in datasets acquired as part of a pragmatic multicentre study of whole body MRI cancer and tested the impact of normalisation methods. We found large variations in ADC within both the same platform, and between different MRI platforms. Normalisation had limited benefit and only for one platform. Fat suppression seemed to be the predominant driver of variation particular in tissues outside the brain.

Background:

Diffusion-Weighted (DW) MRI can be used as a quantitative assessment in oncological imaging alongside conventional MRI1, with the potential to be used as a Quantitative Imaging Biomarker (QIB) where low ADC may correlate with high cellularity associated with tumours. However, there is a lack of data evidencing its accuracy and reproducibility in a wide range of applications, particularly in whole body imaging (WBI). As QIB, it should demonstrate accuracy and reproducibility in lesion detection, utility in critical decision-making and treatment/therapy response in clinical drug trials2–5. Previous studies have shown that ADC reproducibility is known to be affected by scanner type, acquisition parameters6,7 and fat suppression (FS) techniques8. It is important to distinguish fluctuations in ADC from the biologically relevant factors such as the microenvironment variations at cellular level2. Reproducibility can therefore be assisted by ensuring consistency in acquisition methods6,7. However, it may also be possible to further reduce variability by using a reference tissue region to normalise tumour ADC values9,10 as, previously reported.

Purpose:

1) To investigate the effect of a range of acquisition protocols including imaging parameters, FS techniques and MRI vendor on ADC variability in a pragmatic multi-centre whole body MRI trial. 2) To examine the effect of normalisation methods to minimise ADC variability in a range of tissue types.

Method:

Retrospective ROI analysis was carried out on the WB-ADC maps from the STREAMLINE study database13,14 (on-going multicentre studies of WB-MRI staging in lung and colon cancer patients). 77 datasets (55% male) were selected, which was acquired on four different scanners (three vendors) at two field strengths (scanning parameters table1). A research fellow placed spherical ROIs in brain (both sides), bone, muscle, liver and spleen to extract mean ADC values in specific tissue types. The group (by scanner) mean values for each ROI were used to calculate the Coefficient of Variance (CoV) as a measure of reproducibility across scanners and subjects. To investigate the usefulness of normalisation an additional ROI was drawn for each subject and the mean ADC value was extracted and used as reference. Specifically tissues used for normalisation were spleen for the liver, and psoas muscle for bone as previously described9,10. In addition the psoas muscle was also used for liver, and the spleen for bone to investigate the effectiveness of the reference tissue outside the acquisition FOV.

Results:

The CoV across subjects for each tissue type in each scanner with and without regional tissue normalisation is shown in table 2 and the mean and SD of ADC is plotted in figure1. Mean ADC varied according to tissue type across the 4 scanners (Figure1). Coefficient of Variance was lowest with brain and greatest in bone (table2). In particular, mean bone ADC was highest in scanners using STIR FS compared to those using SPIR and SPAIR FS. There was no consistent effect of tesla field strength on tissue ADC. Variation in mean tissue ADC was least when scanner C and D (same vendor, 1.5T and STIR FS) were compared. For 3T platforms (difference in vendor), the range of ADC measurement for each normal tissue was greatest for scanner A, which used SPIR FS compared to scanner B which used SPAIR FS. The CoV was greatest in bone for all 4 scanners. The within FOV normalisation method has no effect in reducing the CoV for 3 of the 4 scanners, but there was some benefit for scanner B. Using a tissue of interest outside the FOV added even more variability.

Discussion:

We have demonstrated significant variability in mean ADC of normal tissues (particularly bone) when measured on different MRI scanners. Furthermore, patient variability differed according to the individual scanner. Reference tissue normalisation (even using reference tissue within FOV) had no benefit in 3 of 4 scanners. Variability can be explained by differences in scanning parameters, vendors and in particular we found some evidence that FS technique influenced variability, particularly for bone. We speculate this may be due to different FS methods (unsuppressed olefinic fat using spectral methods11). SPIR FS (scanner A) seemed to induce greater variation in ADC values12. Additional factors such as differences in vendor coil technology may also have an effect.

Conclusion:

In a non-optimized/standardized multi-centre clinical trial there is substantial variation in ADC values across different individuals within and across scanners as measured by CoV. Varying fat suppression techniques are a likely component of the differences seen between vendors. Normalisation to a reference tissue in post-processing showed limited benefit on normal tissue. In quantitative ADC imaging harmonization of acquisition parameters may be the most effective method for minimising variance.

Acknowledgements

NIHR-Funded research (STREAMLINE Trial)

References

1. Messiou, C. & deSouza, N. M. Diffusion Weighted Magnetic Resonance Imaging of metastatic bone disease: A biomarker for treatment response monitoring. Cancer Biomark. 6, 21–32 (2010).

2. Padhani, A. R. et al. Diffusion-Weighted Magnetic Resonance Imaging as a Cancer Biomarker?: Consensus and Recommendations. 11, 102–125 (2009).

3. Patterson, D. M., Padhani, A. R. & Collins, D. J. Technology Insight: water diffusion MRI—a potential new biomarker of response to cancer therapy. Nat. Clin. Pract. Oncol. 5, 220–233 (2008).

4. Rudin, M. Imaging readouts as biomarkers or surrogate parameters for the assessment of therapeutic interventions. Eur. Radiol. 17, 2441–2457 (2007).

5. Deckers, F. et al. Apparent diffusion coefficient measurements as very early predictive markers of response to chemotherapy in hepatic metastasis: A preliminary investigation of reproducibility and diagnostic value. J. Magn. Reson. Imaging 40, 448–456 (2014).

6. Malyarenko, D. et al. Multi-system repeatability and reproducibility of apparent diffusion coefficient measurement using an ice-water phantom. J. Magn. Reson. Imaging 37, 1238–1246 (2013).

7. Sasaki, M. et al. Variability in absolute apparent diffusion coefficient values across different platforms may be substantial: a multivendor, multi-institutional comparison study. Radiology 249, 624–630 (2008).

8. Poyraz, A. K., Onur, M. R., Kocakoç, E. & Ogur, E. Diffusion-weighted MRI of fatty liver. J. Magn. Reson. Imaging 35, 1108–1111 (2012).

9. Do, R. K. G. et al. Diagnosis of liver fibrosis and cirrhosis with diffusion-weighted imaging: Value of normalized apparent diffusion coefficient using the spleen as reference organ. Am. J. Roentgenol. 195, 671–676 (2010).

10. Padhani, A. R., van Ree, K., Collins, D. J., D’Sa, S. & Makris, A. Assessing the Relation Between Bone Marrow Signal Intensity and Apparent Diffusion Coefficient in Diffusion-Weighted MRI. Am. J. Roentgenol. 200, 163–170 (2013).

11. Abe, T. Frequency-Selective Fat Suppression Radiofrequency Pulse Train to Remove Olefinic Fats. Appl. Magn. Reson. 44, 1213–1221 (2013).

12. Winfield, J. M., Douglas, N. H. M., Desouza, N. M. & Collins, D. J. Phantom for assessment of fat suppression in large field-of-view diffusion-weighted magnetic resonance imaging. Phys. Med. Biol. 59, 2235–48 (2014).

13. http://www.cancerresearchuk.org/about-cancer/find-a-clinical-trial/a-study- looking-mri-scan-diagnose-bowel-cancer-streamline-c

14. http://www.cancerresearchuk.org/about-cancer/find-a-clinical-trial/a-study-looking-mri-scan-diagnose-non-small-cell-lung-cancer-streamline-l

Figures

Table 1 Scanning parameter

Table 2 No normalisation vs normalisation method – Coefficient of Variance (CoV) on different tissue types and scanner types

Figure 1 Box plot: ADC values across scanners and tissue types



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
3878