Fractional Enhancement metric improves SNR and visualisation of quantitative two-point contrast-enhanced MRI in retroperitoneal sarcoma
Matthew David Blackledge1, Christina Messiou1,2, Jessica M Winfield1,2, Dow Mu Koh1,2, David J Collins1,2, and Martin O Leach1,2

1CRUK Cancer Imaging Center, Division of Radiotherapy and Imaging, Institute of Cancer Research, London, United Kingdom, 2MRI, Royal Marsden Hospital, London, United Kingdom

Synopsis

We compare two enhancement fraction parameters that may be used for quantification of two-point contrast-enhanced MRI studies: The relative enhancement and the fractional enhancement. Using computer simulations we show that fractional enhancement is better behaved in the presence of imaging noise, resulting in better SNR for this parameter over a range of intrinsic longitudinal tissue relaxivities and contrast medium concentrations. Further, in a cohort of 25 patients with retroperitoneal sarcoma, fractional enhancement significantly outperformed the relative enhancement in terms of visual assessment of contrast-to-noise, signal-to-noise, tumour detection, imaging artefacts and within tumour contrast.

Background

Volumetric, large field-of-view contrast enhanced MRI typically involves only two MR measurements: One before contrast administration, S1, and another, S2, at some time $$$t$$$ when equilibrium of contrast is expected. To provide quantification of tissue enhancement in these cases, it is common to calculate the relative enhancement, $$$\epsilon_{R}$$$, at each voxel location: $$ \epsilon_{R} = \frac{S_{2} - S_{1}}{S_{1}}$$ Another possibility is to compute the fractional enhancement, $$$\epsilon_{F}$$$: $$ \epsilon_{R} = \frac{S_{2} - S_{1}}{S_{1} + S_{2}}$$

It can be shown that both quantitative parameters are monotonically increasing with increasing contrast agent concentration. Furthermore, both parameters are independent of T2*-weighting, proton density and coil sensitivity, improving their performance as quantitative metrics. However, little work has been done to investigate the statistical properties of these parameters and how this may affect clinical interpretation of contrast enhancement studies.

Purpose

In this article we investigate the contrast-to-noise properties of both the relative and fractional enhancement indices. We demonstrate preliminary evidence that $$$\epsilon_{F}$$$ outperforms $$$\epsilon_{F}$$$ in terms of signal-to-noise ratio (SNR) through numerical simulation, and improves clinical visualization of retroperitoneal sarcoma.

Methods

Patients: Twenty-five patients with retroperitoneal sarcoma were imaged as part of a prospective single-centre study including: 21 well-differentiated/dedifferentiated liposarcomas, 3 leiomyosarcomas and 1 lipoma. All patients provided written consent prior to their involvement in this study.

Imaging: T1-weighted imaging was acquired before and 4 minutes after the administration of gadolinium-based contrast (Dotarem, 0.2 ml/kg boy weight administered at 2 ml/s using a power injector) ensuring that the field of view covered the entire tumour in each patient. We used a 3D FLASH sequence with 17° flip angle ($$$\alpha$$$), repetition time (TR) = 3.8ms and echo time (TE) = 1.06ms on a 1.5T machine (Aera, Siemens Healthcare, Germany).

Simulations: We investigated the response of enhancement parameters, $$$\epsilon_{R}$$$ and $$$\epsilon_{F}$$$, by calculating the expected values as a function of contrast agent, [CA], over the range 0-5 mM in the absence of noise. We use the standard formulae for T1w signal intensity:

$$ \text{S}_{1}(\text{TR}, \alpha) = \text{S}_{0}\sin (\alpha)\frac{1 - \text{E}_{1}}{1-\cos (\alpha)\text{E}_{1}}, \quad \text{E}_{1} = \exp \left\{-\text{TR}\cdot\text{R}_{1}\right\}$$

$$ \text{S}_{2}(\text{TR}, \alpha) = \text{S}_{0}\sin (\alpha)\frac{1 - \text{E}_{1}\text{E}^{\Delta}_{1}}{1-\cos (\alpha)\text{E}_{1}\text{E}^{\Delta}_{1}}, \quad \text{E}^{\Delta}_{1} = \exp \left\{-\text{TR}\cdot\Delta\text{R}_{1}\right\}, \quad\Delta\text{R}_{1} = r_{1}\text{[CA]}$$

We matched all parameters in these simulations with those used for clinical imaging, matching the relaxivity of the contrast agent used ($$$r_{1} = 3.6 \text{ L mmol}^{-1}\text{ s}^{-1}$$$ [1]), over a range of plausible tissue T1-values (0.1-1.5 ms). We repeated these calculations but with the inclusion of Rician noise [2] at different SNRs ($$$\sigma_{0}$$$ = 20, 50, 80, 110, 140 and 170) for the base signal intensity, $$$\text{S}_{0}$$$. This simulation was repeated 105 times so that estimates of SNR for each of the enhancement parameters, $$$\epsilon_{R}$$$ and $$$\epsilon_{F}$$$, could be calculated over a range of T1 and [CA] values.

Image analysis: A clinical radiologist with 15 years experience in MR-imaging compared volumetric datasets of calculated enhancement parameters, $$$\epsilon_{R}$$$ and $$$\epsilon_{F}$$$, in all 25 patients. Images were viewed side-by-side on a multi-planar-reformat workstation (OsiriX, Switzerland); the radiologist was blinded to the method used ($$$\epsilon_{R}$$$ or $$$\epsilon_{F}$$$). For each of the following subjective criteria: (i) Overall contrast-to-noise, (ii) signal-to-noise, (iii) detection of disease, (iv) image artefacts and (v) within-lesion contrast, the radiologist decided whether one method was preferred over the other, or whether they tied.

Results

Figure 1 presents the results from simulations in the absence of noise. It clearly depicts the positive monotonicity of both enhancement parameters as a function of contrast agent concentration. Figure 2 presents the same data but in the presence of Rician noise added to the S0 term, and Figure 3 depicts the SNR of $$$\epsilon_{R}$$$ and $$$\epsilon_{F}$$$ over repeated simulations of Figure 2. Both figures demonstrate that the theoretical noise properties of fractional enhancement $$$\epsilon_{F}$$$ are superior to relative enhancement $$$\epsilon_{R}$$$. Results from the radiological assessment are presented in Figure 4, where it is demonstrated that $$$\epsilon_{F}$$$ outperforms $$$\epsilon_{R}$$$ for all criteria (examples shown in Figure 5).

Discussion and Conclusions

We have investigated the effects of image noise on two possible quantification methods for two-point dynamic contrast enhanced MR measurements, the relative enhancement and fractional enhancement ($$$\epsilon_{R}$$$ and $$$\epsilon_{F}$$$ respectively). Through numerical simulations we have demonstrated that $$$\epsilon_{R}$$$ provides more uniform variation as a function of intrinsic tissue T1 and contrast agent concentration but $$$\epsilon_{F}$$$ outperforms in the presence of image noise. This was validated in a clinical setting of 25 patients with retroperitoneal sarcoma where $$$\epsilon_{F}$$$ was statistically superior in terms of image quality and tumour detection. We thus conclude that $$$\epsilon_{F}$$$ provides a more robust measurement for two-point dynamic MR studies.

Acknowledgements

CRUK and EPSRC support to the Cancer Imaging Centre at ICR and RMH in association with MRC and Department of Health C1060/A10334, C1060/A16464 and NHS funding to the NIHR Biomedical Research Centre and the Clinical Research Facility in Imaging.

References

[1] Rohrer, M., Bauer, H., Mintrovitch, J. et al. “Comparison of Magnetic Properties of MRI Contrast Media Solutions at Different Magnetic Field Strengths”, Investigative Radiology, 40(11), 2005

[2] Gudbjartsson, H. and Patz, S. “The Rician Distribution of Noisy MR data”, Mag. Reson. Med., 34(6), 1995.

Figures

Simulation plots for relative enhancement (left) and fractional enhancement (right) in the absence of noise over a range of contrast agent concentrations and intrinsic tissue T1 values. Note that both indices are monotonically increasing as a function of concentrations and T1. It may be noted that relative enhancement appears to change more uniformly as a function of T1 and concentration.

Simulation plots for relative enhancement (left) and fractional enhancement (right) for six different SNR values of $$$\text{S}_{0}$$$. Note that as $$$\sigma_{0}$$$ becomes low ($$$\sigma_{0}$$$ < 80), the relative enhancement is confounded by the presence of multiple noise outliers, making it difficult to visualise the differences due to intrinsic tissue T1 value and/or contrast agent concentration.

Plots of SNR for relative enhancement (red) and fractional enhancement (green) for six different SNR values for $$$\text{S}_{0}$$$. Note that the SNR of the fractional enhancement outperforms its counterpart in all cases and over all contrast agent concentrations and underlying tissue T1 values.

Results from radiological assessment of relative enhancement fraction $$$\epsilon_{R}$$$ against fractional enhancement $$$\epsilon_{R}$$$ in 25 patients with retroperitoneal sarcoma. Results that are statistically significant (p<0.005, exact multinomial test) are indicated with a double asterisk (**).

Examples of relative enhancement (left) and fractional enhancement maps (right) for two patient examples from the sarcoma study. Note that windowing of the fractional enhancement is much easier due to the absence of possible noise outliers and allows the clinician to visualise the underlying tumour heterogeneity without the presence of signal saturation (red arrow).



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
2454