0595
Influence of residual fat signal on diffusion kurtosis MRI of suspicious mammography findings1Medical Physics in Radiology, German Cancer Research Center (DKFZ), Heidelberg, Germany, 2Radiology, German Cancer Research Center (DKFZ), Heidelberg, Germany, 3Institute of Radiology, University Hospital Erlangen, Erlangen, Germany, 4Radiological Practice at the ATOS Clinic Heidelberg, Heidelberg, Germany, 5Radiology Center Mannheim (RZM), Mannheim, Germany
Synopsis
One of the factors determining the success of diffusion-weighted imaging of the female breast is complete fat suppression, especially when using high b-values. In this study, modified diffusion kurtosis models accounting for residual fat signal were compared to conventional DWI approaches. The comparison was based on a MR mammography dataset acquired in two study centers. The dataset comprised 198 patients with suspicious lesions detected during X-ray mammography screening. The ROC analysis shows significantly better performance of the modified diffusion kurtosis model in discriminating between malignant and benign lesions. This could improve the diagnostic accuracy regarding ambiguous mammography findings.
INTRODUCTION
In the field of the non-invasive methods for characterization of suspicious mammography findings, diffusion-weighted MRI (DWI) gains increasing importance1. In DWI exams of female breast lesions, an efficient fat suppression plays a crucial role, especially at high b-values employed for techniques such as diffusion kurtosis imaging. However, full suppression is often difficult to achieve. Not completely suppressed fat signal can contaminate the signal detected in lesions and lead to distortion of quantitative parameters due to the low diffusion coefficient of fat2. This work aims to evaluate kurtosis fitting models accounting for contamination caused by residual fat signal.METHODS
This retrospective analysis of prospectively acquired data involves datasets of 198 patients with suspicious mammography findings (BI-RADS 4 or 5) who received an indication for breast biopsy during the clarification process. DWI scans were acquired prior to biopsy, in prone position, in one of two study centres (Group A – 1.5T Philips, Ingenia, two-channel loop coil with additive elements on the MRI table; Group B – 1.5T Siemens, Aera, 18-channel breast coil). In Group A (105 patients) a single-shot echo-planar (EPI) sequence was used, whereas in Group B (93 patients) a multishot EPI with three read-out segments was applied. Both sequences were acquired with 4 b-values (0, 100, 750 and 1500 s/mm2) and SPAIR (SPectral Attenuated Inversion Recovery) technique for fat suppression. Detailed parameters of DWI sequence are summarized in the Table 1. Regions of interest (ROIs) were localized and segmented on the slice with the highest b-value where the lesion was visible, based on the T2-weighted image and information included in the X-ray mammography screening report. Fat ROIs were delineated in the region of adipose tissue, usually on the contralateral breast. The following curve-fitting models have been compared:
1) Mono-exponential model: $$S(b)=S_0\cdot{e}^{-b\,\cdot{ADC}}$$ where ADC denotes the apparent diffusion coefficient.
2) Diffusion kurtosis equation: $$S(b)=S_0\cdot{e}^{(-b\,\cdot{ADC}+\frac{1}{6}b^2ADC^2AKC)}$$ where AKC is the apparent kurtosis coefficient.
3) Empirically modified diffusion kurtosis model3, taking into account the contamination from residual fat signal: $$S(b)=\sqrt{\theta(b)^2+\left(S_0\cdot{e}^{(-b\,\cdot{ADC}+\frac{1}{6}b^2ADC^2AKC)}\right)^2}$$ where θ(b) is the signal intensity in the fatty tissue ROI on the contralateral side.
4) Altered kurtosis model, including fat signal: $$S(b)=S_0\cdot{e}^{(-b\,\cdot{ADC}+\frac{1}{6}b^2ADC^2AKC)}+\theta(b).$$ As a rough measure of conspicuity of the lesion, the ratio of the mean signal intensity in the lesion to the mean intensity in the fatty area on the b-value=1500 s/mm² images was used.
RESULTS
Using the kurtosis approach, the values of ADC and AKC were used as features in a logistic regression model allowing for prediction of malignancy of the lesion. In contrast, in case of Model1, the ADC was the only available feature. The analysis of receiver operating characteristic (ROC) curves (Figure 2) showed that the values of area under the curve (AUC) obtained for Model3 [0.86 (95% CI 0.80-0.91)] and Model4 [0.86 (95% CI 0.80-0.91)] were significantly higher (p< 0.03) than AUC values of Model1 [0.77 (95% CI 0.70-0.84)] and Model2 [0.79 (95% CI 0.72-0.86)]. Considering individually each group of patients, in Group B the differences between models were not statistically significant (p-values for pairwise comparison between all models were higher than 0.06), while in Group A Model3 performed significantly better than Model1 and Model2 (p≤ 0.02). Dividing each cohort of patients into three subgroups according to the ratio of the mean signals in the lesion and in the fatty area revealed that Model3 presents the best performance among all methods in both study sites in the subgroup with the lowest ratio.DISCUSSION
Due to the low diffusion coefficient of fat, incomplete fat suppression can form a considerable background signal level. This may contaminate the signal obtained in lesion ROIs by partial volume effects or by chemical shift or ghosting, depending on the exact sequence parameters, fat suppression techniques and post-processing methods. Thus, a phenomenological extension was used here to reduce the effect of residual fat signal on diffusion kurtosis parameters. This yielded an improvement, especially for one study center, emphasizing the dependence of the effect on the specific MR imaging parameters.CONCLUSION
The extension of diffusion kurtosis models by terms accounting for residual fat signal can improve the diagnostic accuracy of DWI of suspicious lesions in the female breast, which can be useful especially in case of the inadequate fat suppression. In the case of lower fat contamination, the proposed approach does not result in adverse effects, which suggests a wide applicability of the method.
Acknowledgements
No acknowledgement found.References
1. Bickelhaupt S, Laun FB, Tesdorff J, Lederer W, Daniel H, Stieber A, Delorme S, Schlemmer HP. Fast and Noninvasive Characterization of Suspicious Lesions Detected at Breast Cancer X-Ray Screening: Capability of Diffusion-weighted MR Imaging with MIPs. Radiology. 2016 Mar.278(3):689-97.
2. Baron P, Dorrius MD, Kappert P, Oudkerk M, Sijens PE. Diffusion-Weighted Imaging of Normal Fibroglandular Breast Tissue: Influence of Microperfusion and Fat Suppression Technique on the Apparent Diffusion Coefficient. NMR in Biomedicine. 23.4 (2010). 399–405.
3. Jensen JH, Helpern JA, Ramani A, Lu H, Kaczynski K. Diffusional Kurtosis Imaging: The Quantification of Non-Gaussian Water Diffusion by Means of Magnetic Resonance Imaging. Magn Reson Med. 2005 Jun;53(6):1432-40.
Figures
