The test-retest reliability of fat-water ratio MRI derived breast density measurements and automated breast segmentation
Jie Ding1, Patricia A Thompson2, Marilyn T Marron3, Maria Altbach3,4, Denise Roe3,5, Jean-Philippe Galons4, Cynthia A Thomson3, Fang Wang6, Alison Stopeck7, and Chuan Huang1,8,9

1Biomedical Engineering, Stony Brook University, Stony Brook, NY, United States, 2Pathology, Stony Brook Medicine, Stony Brook, NY, United States, 3Cancer Center, University of Arizona, Tucson, AZ, United States, 4Medical Imaging, University of Arizona, Tucson, AZ, United States, 5Epidemiology and Biostatistics, University of Arizona, Tucson, AZ, United States, 6Stony Brook Medicine, Stony Brook, NY, United States, 7Hematology and Oncology, Stony Brook Medicine, Stony Brook, NY, United States, 8Radiology, Stony Brook Medicine, Stony Brook, NY, United States, 9Psychiatry, Stony Brook Medicine, Stony Brook, NY, United States

Synopsis

It has been shown that breast density (BD) value derived from fat-water-ratio MRI (FWR-MRI) strongly correlates with standard digital mammogram derived BD. The fact that no ionizing radiation is associated with FWR-MRI makes it a lower-risk modality for long term BD monitoring and clinical trials. However, data regarding the individual and group level variability and reliability of this method needs to be established. Conventional approaches for FWR-MRI derived BD rely on manually drawn regions-of-interest. These processes are cumbersome and prone to measurement bias, which may limit the application of FWR-MRI derived BD. Automated breast segmentation has been proposed to resolve this problem and limited results to date are promising. Additional data including an evaluation of BD reliability from manual versus automated measurements is still needed. In this study, we evaluate the test-retest reliability of the FWR-MRI derived BD and the quality of data using manual versus automated breast segmentation. Our results demonstrate the high reliability of the FWR-MRI derived BD measure, Fra80, with a typical error of less than 0.02 for both automated and manual breast segmentation. Moreover, our automated breast segmentation protocol yields more reliable Fra80 BD measures compared to the labor-intense manual segmentation method.

Purpose

It has been shown that breast density (BD) value derived from fat-water-ratio MRI (FWR-MRI) strongly correlates with standard digital mammogram derived BD [1]. The fact that no ionizing radiation is associated with FWR-MRI makes it a lower-risk modality for long term BD monitoring and clinical trials. However, data regarding the individual and group level variability and reliability of this method needs to be established.

Conventional approaches for FWR-MRI derived BD rely on manually drawn regions-of-interest. These processes are cumbersome and prone to measurement bias, which may limit the application of FWR-MRI derived BD. Automated breast segmentation has been proposed to resolve this problem and limited results to date are promising [2]. Additional data including an evaluation of BD reliability from manual versus automated measurements is still needed.

In this study, we evaluate the test-retest reliability of the FWR-MRI derived BD and the quality of data using manual versus automated breast segmentation.

Methods

In an ongoing prevention trial, we have identified 8 participants who provided repeat FWR-MRI scans over 1 year (5 participants with 3 repeated scans, 2 with 2 repeated scans, 1 with 1 repeated scans). For each repeated scan, the patient was taken out of the scanner and put back in immediately, re-registered and re-localized. FWR-MRI scans were performed on a 1.5T GE Signa NV-CV/i scanner using an axial radial GRASE acquisition [3]. Fat-water separation was performed using a region growing IDEAL technique optimized for the radial GRASE sequence [3, 4]. Breast segmentation was initially performed manually (20-35 minutes per volume), followed by automated breast segmentation which was applied to all 40 (20×2) scans using a technique similar to Ref. 2 but with improved 3D regularization. Manual segmentation was performed according to Ref. 2.

The FWR-MRI derived BD was calculated from quantification of the ratio of breast voxels with less than 80% fat fraction (Fra80, range 0-1, higher number indicated denser breast) in the entire breast. This measure has been previously shown to be strongly correlated with mammogram based BD measurement [5].

Test-retest reliability was evaluated statistically using within-subject standard deviation (aka typical error) and intra-class correlation (ICC) analyses in Matlab. ICC [6] was performed on the logarithm of the Fra80. All enrolled participants had prior early stage breast cancer and thus only the contralateral, unaffected breast was analyzed.

Results and Discussion

A representative slice of FWR-MRI images is shown in Figure 1, along with its manual segmentation and automated segmentation results. As shown, manual and automated segmentation have excellent agreement visually.

Figure 2 shows the test-retest reliability of Fra80 using either manual segmentation or automated segmentation. Fra80 as quantified by either manual or automated segmentation yields very similar and reproducible results. The typical error for manual segmentation and automated segmentation are 0.0192 and 0.0135, respectively; and the corresponding ICC are 0.9534 (95% confidence interval = 0.8883-0.9812) and 0.9814 (95% confidence interval = 0.9544-0.9925), respectively.

Conclusion

Our results demonstrate the high reliability of the FWR-MRI derived BD measure, Fra80, with a typical error of less than 0.02 for both automated and manual breast segmentation. Moreover, our automated breast segmentation protocol yields more reliable Fra80 BD measures compared to the labor-intense manual segmentation method.

Acknowledgements

This work is partially supported by NIH grants CA149417, CA161534.

References

1. Trouard T, et al, Fat water ratio and diffusion-weighted MRI applied to themeasure of breast density as a cancer risk biomarker, Proc. ISMRM 18 (2010) 4749

2. Rosado-Toro J, et al. Automated breast segmentation of fat and water MR images using dynamic programming. Academic radiology 22.2 (2015): pp. 139-148

3. Li Z, et al. Rapid water and lipid imaging with T2 mapping using a radial IDEAL-GRASE technique. Magnet Reson Med, 6 (2009), pp. 1415-1424

4. Huang C, Altbach MI. Multi-Mask Multi-Seed Free Growing Field Map Estimation Algorithm for Iterative Multi-Point Water-Fat Decomposition. Proc. ISMRM 17 (2009) 2809

5. Thomson C, et al. 2-Hydroxyestrone is associated with breast density measured by mammography and fat:water ratio magnetic resonance imaging in women taking tamoxifen [abstract]. In: Proceedings of the Thirty-Seventh Annual CTRC-AACR San Antonio Breast Cancer Symposium: 2014 Dec 9-13; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2015;75(9 Suppl):Abstract nr P6-01-18

6. McGraw, K. O., Wong, S. P., Forming Inferences About Some Intraclass Correlation Coefficients. Psychological Methods, Vol. 1, No. 1, pp. 30-46, 1996

Figures

Figure 1. (left) A representative breast fat fraction map, and its corresponding (center) manual and (right) automated segmentation results.

Figure 2. The test-retest reliability of Fra80 using either (left) manual segmentation or (right) automated segmentation. It can be seen that Fra80 with both manual and automated segmentation yields very reproducible results. The typical error for manual segmentation and automated segmentation are 0.0192 and 0.0135, respectively; and the corresponding ICC are 0.9019 and 0.9645, respectively.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
2472