DCE-MRI: Analysis
Lucy Elizabeth Kershaw1
1The University of Edinburgh, Edinburgh, United Kingdom

Synopsis

Analysing DCE-MRI data can be time-consuming and complex. In this session, the process will be broken down into steps whilst highlighting potential pitfalls. Analysis and acquisition are closely linked but we will start from the position of having acquired anatomical images, T1 maps and the dynamic series, and assume that analysis will be done using python, matlab etc. We will then cover:

  • AIF extraction
  • Signal intensity to contrast agent concentration conversion
  • Bolus arrival time determination
  • Model selection
  • Model fitting
  • Alternative simple analysis options

Objective

To provide a practical understanding of the different steps involved in analysis of DCE-MRI data, suitable for basic research scientists and clinicians with an interest in python/matlab etc

Course Summary

This session will start from the assumption that the following has been acquired:
  1. Anatomical reference images
  2. T1 mapping images (or calculated map)
  3. Dynamic series
The critical first step is to look at the data as part of a basic quality control check. Has there been excessive motion? Do the T1 values look sensible? Are there any image artefacts (e.g. from wrap, parallel imaging, motion or flow?) Has there been any enhancement in the tissue of interest? Is there a clear feeding vessel present for AIF measurement? If there are issues with the images, consider how to alter the acquisition protocol to improve things for the next study - acquisition and analysis go hand-in-hand.

Choosing an AIF is notoriously difficult and there is no gold-standard way to do this. We will examine the potential pitfalls with AIF selection and what effect this has on the final results, as well as discussing the alternative option of a population AIF (1).

Signal intensity is generally converted to contrast agent concentration via the baseline T1 measurement (2), but this can introduce errors, particularly in conversion of the AIF. The conversion can be done as part of the model fitting or in advance, both of which have advantages and disadvantages.

Bolus arrival time is the time taken for blood to travel from the feeding vessel where the AIF is measured to the tissue of interest, and should be taken account of in the analysis (3). Again, no gold standard method has emerged but this can be done as part of the model fitting (4) or as a separate step (5).

A detailed look at model selection and fitting is beyond the scope of this session but the basic principles will be discussed, including a general description of commonly-used model families (6, 7) along with special cases such as the liver and kidneys. Model fitting requires a tailored approach depending on the particular dataset and number of free parameters to be included, but in general nonlinear model fitting will require keeping track of fit failures and taking steps to ensure a global rather than local minimum is found (8). Linearisation of some models (9) may result in fit improvements, and the application of deep learning may also be advantageous (10, 11). Parallelisation can lead to major reductions in fitting time, but often requires input from specialists. A recent initiative that aims to build a python library for DCE-MRI, DSC-MRI and ASL analysis may be a useful starting point for those aiming to develop their own analysis pipeline [https://www.osipi.org].

Reporting parameters from DCE-MRI analysis also requires careful consideration. A mean or median from a region of interest is simple, but risks throwing away useful spatial information and risks bias in region selection. When reporting changes in parameters over time, how can it be ensured that the same tissue is being compared? If the model fit is good for one time point but poor for another, how should this be reported? Finally, we will cover simpler methods for data analysis that may provide useful parameters (12).

Acknowledgements

No acknowledgement found.

References

1. Parker GJM, Roberts C, Macdonald A, et al.: Experimentally-derived functional form for a population-averaged high-temporal-resolution arterial input function for dynamic contrast-enhanced MRI. Magn Reson Med 2006; 56:993–1000.

2. Brookes JA, Redpath TW, Gilbert FJ, Murray AD, Staff RT: Accuracy of T1 measurement in dynamic contrast-enhanced breast MRI using two- and three-dimensional variable flip angle fast low-angle shot. J Magn Reson Imaging 1999; 9:163–171.

3. Mehrtash A, Gupta SN, Shanbhag D, et al.: Bolus arrival time and its effect on tissue characterization with dynamic contrast-enhanced magnetic resonance imaging. J Med Imaging 2016; 3:014503.

4. Kershaw LE, Buckley DL: Precision in measurements of perfusion and microvascular permeability with T(1)-weighted dynamic contrast-enhanced MRI. Magn Reson Med 2006; 56:986–992.

5. Bendinger AL, Debus C, Glowa C, Karger CP, Peter J, Storath M: Bolus arrival time estimation in dynamic contrast-enhanced magnetic resonance imaging of small animals based on spline models. Phys Med Biol2019; 64:045003.

6. Sourbron SP, Buckley DL: Tracer kinetic modelling in MRI: estimating perfusion and capillary permeability. Phys Med Biol 2012; 57:R1-33.

7. Sourbron SP, Buckley DL: Classic models for dynamic contrast-enhanced MRI. NMR Biomed 2013; 26:1004–27.

8. Motulsky H, Christopoulos A: Fitting Models to Biological Data Using Linear and Nonlinear Regression: A Practical Guide to Curve Fitting. OUP; 2004.

9. Flouri D, Lesnic D, Sourbron SP: Fitting the two-compartment model in DCE-MRI by linear inversion. Magn Reson Med 2016; 76:998–1006.

10. Ulas C, Das D, Thrippleton MJ, et al.: Convolutional Neural Networks for Direct Inference of Pharmacokinetic Parameters: Application to Stroke Dynamic Contrast-Enhanced MRI. Front Neurol 2019; 9(January):1–14.

11. Strzelecki M, Klepaczko A, Muszelska M, Eikeford E, Rorvik J, Lundervold A: An artificial neural network for GFR estimation in the DCE-MRI studies of the kidneys. Signal Process - Algorithms, Archit Arrange Appl Conf Proceedings, SPA 2018; 2018-Septe:286–291.

12. Dijkhoff RAP, Maas M, Martens MH, et al.: Correlation between quantitative and semiquantitative parameters in DCE-MRI with a blood pool agent in rectal cancer: can semiquantitative parameters be used as a surrogate for quantitative parameters? Abdom Radiol 2017; 42:1342–1349

Proc. Intl. Soc. Mag. Reson. Med. 29 (2021)