4316

Optimizing b-value Sampling Strategies with Integrative DWI: A Comparison Study on DISMANTLE Parameters in Breast

Jacqueline Escutia¹, Guangyu Dan^1,2, Albert Yen³, Erin Neuschler⁴, Xiaohong Joe Zhou^1,2,4,5, and Muge Karaman^1,2
¹Department of Biomedical Engineering, University of Illinois Chicago, Chicago, IL, United States, ²Center for MR Research, University of Illinois Chicago, Chicago, IL, United States, ³University of Illinois Chicago, Chicago, IL, United States, ⁴Department of Radiology, University of Illinois Chicago, Chicago, IL, United States, ⁵Department of Neurosurgery, University of Illinois Chicago, Chicago, IL, United States

Synopsis

Keywords: Breast, Breast, Tissue heterogeneity, tissue vascularity, perfusion, advanced DWI, non-Gaussian DWI

Motivation: This study is driven by the pressing need for the standardization of breast diffusion-weighted imaging (DWI) techniques.

Goal(s): The objective is to explore the influence of b-value selection on the parameters estimated by using an integrated DWI approach (DISMANTLE) that aims to simultaneously assess tissue cellularity, vascularity, and heterogeneity.

Approach: We conducted a systematic analysis by evaluating DISMANTLE parameters in healthy breast tissue using diverse b-value sampling strategies.

Results: Our findings identified an ideal b-value distribution for accurate implementation of DISMANTLE in breast. This dataset included 12 b-values, reducing scan time by 34% compared to the comprehensive DWI protocol featuring broad b-value range.

Impact: This study emphasizes the role of b-value sampling in advanced breast DWI. Our systematic evaluation contributes to the potential success of advanced DWI techniques in breast imaging and the current discussion of the need for standardization in advanced DWI.

Introduction

Accurate in vivo characterization of breast tissue is important for efficient risk assessment, diagnosis, and optimized treatment strategies in breast cancer. Diffusion-weighted imaging (DWI) with apparent diffusion coefficient has become a pillar of clinical MRI for probing tissue cellularity¹. Recent advancements in DWI have revealed that specific ranges of b-values in the diffusion-weighted signal can provide insights into various tissue properties^2,3. One of these advanced techniques is an integrated DWI approach for simultaneous assessment of tissue cellularity, vascularity, and heterogeneity – so called DISMANTLE⁴ – which combines the intravoxel incoherent motion (IVIM) model⁵ at low b-values with a high-b-value continuous-time random-walk (CTRW) model⁶. As emphasized by the EUSOBI International Breast DWI working group⁷, the challenge in advanced breast DWI lies in standardization where the number of b-values and the choice of their distribution plays a crucial role. Considering that different ranges of b-value selections sensitize the signal to different components of diffusion, this decision can make a substantial difference in the parameters, particularly those recognized for their variability⁸. In this study, we performed a systematic comparison among the DISMANTLE parameters in the breast tissue of a healthy volunteer obtained from a total of eight different b-value sampling strategies extracted from a comprehensive DWI protocol with a wide b-value range.

Methods

Datasets: A healthy volunteer underwent an axial MRI scan at 3T (GE Healthcare, Discovery MR750) with an 8-channel breast coil. DWI was performed with 15 b-values (0-3000 s/mm²), TR/TE=7000/78ms, slice thickness=5mm, FOV=32cm×32cm, and matrix=256×256). This comprehensive dataset is referred to as “Dataset_Comp” from which an additional seven subsets, Dataset_A-G, were generated. The datasets and corresponding scan times are illustrated in Figs. 1a and 1b, respectively.
DWI Image Analysis: The diffusion-weighted images in all datasets were analyzed using DISMANTLE⁴
$$ S/S_0=fe(-bD_{perf } )+(1-f) E_α (-(bD_m )^β), (1) $$
where E_α is a Mittag-Leffler function. Eq.(1) produces vascularity-related (IVIM’s perfusion fraction, f and pseudo-diffusion coefficient, D_perf), cellularity-related (CTRW’s diffusion coefficient, D_m), and heterogeneity-related (CTRW’s temporal and spatial diffusion heterogeneity, α and β) parameters. DISMANTLE was implemented through a multi-step approach as in Figure 1c.
Statistical Analysis: A total of 100 regions of interest (ROIs) with 5×5 voxels were drawn on five consecutive mid-slices of breast parenchyma. The mean values of DISMANTLE parameters were computed over the ROIs. Comparison of mean parameter values from each of the datasets was performed using a Mann-Whitney-U test (significance threshold: 0.05); and the results were demonstrated using a heatmap. Coefficients of variation (CV) were calculated as the ratio of standard deviation over the mean for each parameter.

Results

Figures 2a-2e show DISMANTLE parameters obtained from Dataset_Comp and their absolute difference maps between those obtained from Dataset_A, DatasetB, DatasetE, and Dataset_G, respectively, as examples. The parameters are also visualized in Figures 3a and 3b with histograms and boxplots; and summarized with descriptive statistics in Figure 3c. While variations can be observed in most parameters among most datasets, the most striking differences were observed with Dataset_A. Figure 4 shows that statistically significant differences were determined between Dataset_Comp and other datasets in most parameters, with 𝛼 and β exhibiting more pronounced disparities. Figures 5a and 5b show the boxplots and the descriptive statistics of the CV values of each DISMANTLE parameter from each dataset. When the results in Figures 3-5 were evaluated collectively, as explained in figure captions, Dataset_B, Dataset_E and Dataset_G stood out by producing similar parameter values to Dataset_Comp (darker blue cells in heatmap in Figure 4) with comparable or lower CVs and ~6%, and ~12%, ~34% reduction in scan time, respectively (Figure 1b).

Discussion and Conclusion

We systematically compared DISMANTLE parameters within the breast tissue of a healthy volunteer to identify an optimized b-value sampling scheme for advanced DWI analysis. This entailed analyzing multiple datasets derived from a comprehensive b-value sampling scheme while considering parameters' deviation from those obtained from the comprehensive dataset and a low coefficient of variability. Our findings highlighted the effectiveness of Dataset_G, which includes 12 b-values, allowing for a clinically feasible scan time of approximately 10 minutes. Notably, our study revealed that a b-value of 2500 s/mm² suffices for accurate CTRW model fitting in breast tissue, in contrast to high b-value DWI studies in brain³ where b-values can reach 4000 s/mm². This underscores the need for organ-specific DWI protocols and the complexities in selecting b-value schemes for advanced DWI protocols. Our systematic evaluation, with potential expansion to larger sample sizes and diverse disease conditions, contributes to the potential success of advanced breast DWI techniques like DISMANTLE in clinical studies and the current discussion of the need for standardization in advanced DWI.

Acknowledgements

This work was supported in part by the NIH (Grant No. 1R21EB032071).

References

[1] Jiang R, Ma Z, Dong H, et al. Diffusion tensor imaging of breast lesions: Evaluation of apparent diffusion coefficient and fractional anisotropy and tissue cellularity. Br J Radiol. 2016;89(1064).

[2] Le Bihan D. Apparent diffusion coefficient and beyond: What diffusion MR imaging can tell us about tissue structure. Radiology. 2013;268(2):318-322.

[3] Tang L, Zhou XJ. Diffusion MRI of cancer: From low to high b-values. J Magn Reson Imaging. 2019;49(1):23-40.

[4] Karaman M, Che S, Dan G, et al. Integrated DWI for pre-treatment prediction of response to neoadjuvant chemotherapy in breast cancer. Proc ISMRM Annual Meeting; 2022;S401.

[5] Le Bihan D, Breton E, Lallemand D, et al. MR imaging of intravoxel incoherent motions: application to diffusion and perfusion in neurologic disorders. Radiology. 1986;161(2):401-407.

[6] Karaman MM, Sui Y, Wang H, et al. Differentiating low- and high-grade pediatric brain tumors using a continuous-time random-walk diffusion model at high b-values. Magn Reson Med. 2016;76(4):1149-1157.

[7] Baltzer P, Mann RM, Iima M, et al. EUSOBI international Breast Diffusion-Weighted Imaging working group. Diffusion-weighted imaging of the breast-a consensus and mission statement from the EUSOBI International Breast Diffusion-Weighted Imaging working group. Eur Radiol. 2020 Mar;30(3):1436-1450.

[8] Chabert S, Verdu J, Huerta G, et al. Impact of b-Value Sampling Scheme on Brain IVIM Parameter Estimation in Healthy Subjects. Magn Reson Med Sci. 2020 Aug 3;19(3):216-226.

[9] Sigmund EE, Cho GY, Kim S, et al. Intravoxel incoherent motion imaging of tumor microenvironment in locally advanced breast cancer. Magn Reson Med. 2011;65:1437-1447.

[10] Cho GY, Moy L, Zhang JL, et al. Comparison of fitting methods and b-value sampling strategies for intravoxel incoherent motion in breast cancer. Magn Reson Med. 2015;74(4):1077-1085.

Figures

Figure 1: a) An illustration of the b-value distribution of the datasets. Subscripts in the b-values denote the NEXs. b) Acquisition times of each dataset in minutes:seconds. c) Implementation steps of DISMANTLE. Step 1: DWI data in the low-b-value range (0-b_low-max s/mm²) was analyzed with an IVIM model⁹ using segmented fitting¹⁰; Step 2: perfusion-related signal was removed from the diffusion-weighted signal; Step 3: remaining diffusion-related signal component was analyzed over the entire-b-value range (0-b_high-max s/mm²; b_high-max) with CTRW model⁶.

Figure 2: DISMANTLE parameter maps, D_m, α, β, D_perf, and f from Dataset_Comp in a), and the absolute difference maps between the DISMANTLE parameters from Dataset_Comp and a selection of datasets as follows: b) Dataset_A, c) Dataset_B, d) Dataset_E, and e) Dataset_G whose b-value distributions are illustrated in Fig 1a.

Figure 3: a) Boxplots and b) histograms of the DISMANTLE parameters estimated from each of the datasets. The corresponding descriptive statistics, showing sample mean and standard deviation, (x ̅±σ), of each parameter are given in c). Dataset_A, highlighted in a red box, exhibited the most substantial deviation in parameter values compared to Dataset_Comp, while Dataset_B, Dataset_E, and Dataset_G, outlined in green boxes, were the top three datasets that yielded parameter values similar to those of Dataset_Comp.

Figure 4: A heat map of the p-values for the statistical difference between each of the DISMANTLE parameters (column) estimated from the comprehensive dataset, Dataset_Comp, and the extracted subsets, Dataset_A-G (row). Each cell in the heatmap is labeled with a p-value. Values were represented using a spectrum of low (light blue) to high (dark blue). For instance, the p-value at the intersection of the first row (Dataset_A) and first column (D_m) assesses the significance of the difference in D_m between Dataset_Comp and Dataset_A.

Figure 5: a) Boxplots and b) the corresponding descriptive statistics, showing sample mean and standard deviation, (x ̅±σ), of the CVs of each DISMANTLE parameter estimated from each of the datasets. Dataset_A, highlighted in a red box, exhibited the highest CVs across all parameters, aligning with the findings in Fig 3c. On the other hand, Dataset_G, highlighted in green, produced the most favorable CV values among the three datasets marked with green boxes in Fig. 3c.

Proc. Intl. Soc. Mag. Reson. Med. 32 (2024)

4316

DOI: https://doi.org/10.58530/2024/4316