0139

A Hybrid DWI Approach for Simultaneous Assessment of Cellularity, Vascularity, and Heterogeneity of Breast Lesions

Muge Karaman^1,2, Yangyang Bu^3,4, Guangyu Dan^1,2, Zheng Zhong^1,2, Qingfei Luo¹, Shiwei Wang^3,4, Changyu Zhou^3,4, Weihong Hu^3,4, X. Joe Zhou^1,2,5, and Maosheng Xu^3,4
¹Center for Magnetic Resonance Research, University of Illinois at Chicago, Chicago, IL, United States, ²Department of Bioengineering, University of Illinois at Chicago, Chicago, IL, United States, ³The First Clinical Medical College, Zhejiang Chinese Medical University, Hangzhou, China, ⁴Department of Radiology, The First Affiliated Hospital of Zhejiang Chinese Medical University, Hangzhou, China, ⁵Departments of Radiology and Neurosurgery, University of Illinois at Chicago, Chicago, IL, United States

Synopsis

Breast cancer is the second most common cancer among women. Breast tissue has a variety of structural features leading to different tissue properties which are collectively linked to the pathologic state of the breast lesions. In this study we demonstrate a hybrid DWI approach for simultaneous assessment of tissue cellularity, vascularity, and heterogeneity (DISMANTLE) based on intravoxel incoherent motion (IVIM) and continuous-time random walk (CTRW) diffusion models. Our results have shown that DISMANTLE improved the differentiation among benign and malignant breast lesions compared to the classical implementation of either IVIM or CTRW model.

Introduction

Accurate characterization of breast lesions is important for efficient risk assessment and optimized treatment for breast cancer. Breast tissue has various structural features leading to different tissue properties. Among these, changes in cellularity, vascularity, and heterogeneity have been increasingly linked to the pathologic state of lesions. While DWI with apparent diffusion coefficient has become a pillar of clinical MRI for probing tissue callularity^1,2, recent publications indicate that multiple tissue properties can be investigated using specific b-value ranges of the diffusion-weighted signal attenuation which exhibits both Gaussian and non-Gaussian diffusion behaviors^3,4. For example, intravoxel incoherent motion (IVIM) model⁵ with low b-values can characterize micro-vascularity while high-b-value non-Gaussian models can probe microstructures as in the case with continuous-time random-walk (CTRW) model^6,7 which can reflect micro-structural heterogeneity. Although IVIM model has been increasingly used for breast tissue characterization⁸, there is a paucity of high-b-value DWI studies on breast cancer. More importantly, integration of information from a full-b-value “spectrum” would likely lead to new insights beyond what each individual model can provide. In this study, we propose a hybrid DWI approach for simultaneous assessment of tissue cellularity, vascularity, and heterogeneity (DISMANTLE) based on IVIM and CTRW diffusion models to characterize malignant and benign breast lesions.

Theory

The DWI signal can be characterized as the sum of perfusion- and diffusion-related signal components, F_PERF(b) and F_DIFF(b)⁹, $$ S(b)=S_{0,PERF} F_{PERF} (b)+S_{0,DIFF} F_{DIFF} (b).\tag{1}$$ By adopting this hybrid approach, we characterize the diffusion-weighted signal attenuation by approximating F_PERF and F_DIFF based on IVIM^5,10 and CTRW⁷ models, respectively, as follows $$S/S_0=fe^{-bD_{perf}}+(1-f) E_α (-(bD_m )^β),\tag{2}$$ where E_α is a Mittag-Leffler function⁷. The approach in Eq.(2), which we call DISMANTLE, enables estimation of 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 with a unified formulism.

Methods

Image Acquisition: We recruited 33 women with histologically confirmed breast lesions (21 malignant, 12 benign). All patients underwent axial MRI scans at 3T (GE Healthcare, Discovery MR750) with an 8-channel breast coil. DWI was performed with 11 b-values (0₁, 50₁, 100₂, 300₂, 500₂, 800₄, 1100₄, 1500₆, 2000₆, 2500₈, 3000₈ s/mm²; subscripts denoting NEXs; TR/TE=7000/78ms, slice thickness=5mm, FOV=32cm×32cm, and matrix=256×256). Trace-weighted images were obtained to minimize the diffusion anisotropy effect.
DWI Analysis: The diffusion-weighted images were analyzed through a multi-segmented implementation of DISMANTLE with the following steps (Fig.1): Step 1) The images in the low-b-value range (0-800 s/mm²) were first analyzed with IVIM model⁵, $$S/S_0=fe^{(-b(D_{diff}+D_{perf} )) }+(1-f) e^{(-bD_{diff} ) }\tag{3}$$ to estimate f, D_diff, and D_perf using a segmented fitting algorithm¹¹ by first estimating D_diff with a mono-exponential model at mid-range-b-values (300-800 s/mm²), followed by extrapolating the mono-exponential fit to b=0 to estimate f, and constraining D_diff and f in Eq.(3) to obtain D_perf. Step 2) Based on the estimated f and D_perf in Step 1), the perfusion-related signal was calculated according to Eq.(2), and removed from the diffusion-weighted signal. Step 3) The remaining diffusion-related signal component at the entire-b-value range (0-3000 s/mm²) was analyzed with CTRW model⁷, $$S/S_0=E_α (-(bD_m )^β).\tag{4}$$ For comparison, a classical CTRW-based analysis was performed by repeating Step 3) with the use of all b-values without subtracting the perfusion-related signal component.
Statistical Analysis: The mean values of DISMANTLE-based parameters, D_perf, f, D_m, α, and β, were computed over the lesion regions of interest (ROIs). A Mann-Whitney U-test was used for the group comparison, followed by a receiver operating characteristic (ROC) analysis for differentiation among benign and malignant lesions using different combinations of the parameters with a multivariate logistic regression algorithm. For comparison, this process was repeated for two additional scenarios using:1) classical IVIM-based parameters, D_perf, D_diff, f, and 2) classical CTRW-based parameters, D_m, α, and β.

Results

Figures 2a and 2b show parameters from the classical CTRW-based and DISMANTLE-based analysis for a representative benign (first column) and malignant (second column) lesion. D_m values were lower while α and β were higher in the DISMANTLE-based analysis compared to the classical CTRW-based analysis in both patients. The malignant lesion exhibited lower values in D_m and f; and higher values in α, β, and D_perf. While f has typically been found to be higher in malignant lesions, a significant overlap was also reported in previous studies especially when complex benign fibroadenomas with complicated cysts are involved¹². In the group comparison, classical CTRW-based analysis revealed statistically significant differences (p<0.05) only in D_m, whereas all the DISMANTLE-based parameters except D_perf showed significance (Fig.3). In differentiation among the lesions, combination of DISMANTLE-based parameters produced the highest sensitivity (0.916 vs. 0.833 or 0750), specificity (0.950 vs. 0.900 or 0.900), accuracy (0.950 vs. 0.875 or 0.843), and area-under-the-curve (0.912 vs. 0.816 or 0.837) compared to the classical CTRW-based or IVIM-based parameter combinations (Fig.4).

Discussion and Conclusion

We have proposed a hybrid approach that enables a comprehensive characterization of breast tissue to collectively probe tissue cellularity, vascularity, and heterogeneity from a full-b-value spectrum. Using a single representation based on a low-to-moderate-b-value IVIM and a high-b-value CTRW model, DISMANTLE improved characterization of benign and malignant breast lesions compared to the classical implementation of the models. This study emphasizes the importance of integrating information from advanced DWI for improved breast cancer diagnosis.

Acknowledgements

No acknowledgement found.

References

[1] Chen L, Liu M, Bao J, et al. The correlation between apparent diffusion coefficient and tumor cellularity in patients: A meta-analysis. PLoS One. 2013;8(11).

[2] 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).

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

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

[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] Ingo C, Magin RL, Colon-Perez L, Triplett W, Mareci TH. On random walks and entropy in diffusion‐weighted magnetic resonance imaging studies of neural tissue. Magn Reson Med. 2014;71:617–627.

[7] 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.

[8] Partridge SC, Nissan N, Rahbar H, et al. Diffusion-weighted breast MRI: Clinical applications and emerging techniques. J Magn Reson Imaging. 2017;45(2):337-355.

[9] Le Bihan D, Breton E, Lallemand D, et al. Separation of diffusion and perfusion in intravoxel incoherent motion MR imaging. Radiology. 1988: 168(2):497-505.

[10] 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.

[11] 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.

[12] Iima M and Le Bihan D. Clinical intravoxel incoherent motion and diffusion MR imaging: Past, present, and future. Radiology. 2016;278(1):13-32.

Figures

Figure 1: The steps involved in DISMANTLE. Step 1: DWI data in the low-b-value range (0-800 s/mm²) were analyzed with an IVIM model. Step 2: The perfusion-related DWI signal was computed based on estimated D_perf and f from Step 1; and subtracted from the diffusion-weighted signal across the entire-b-value range (0-3000 s/mm²). Step 3: The remaining diffusion-related signal component in the entire b-value range was analyzed with the CTRW model.

Figure 2: Maps of classical CTRW-based parameters, D_m, α, and β in a) and of DISMANTLE-based parameters, D_diff, D_perf, f, D_m, α, and β in b) from representative patients in the benign (first column) and malignant (second column) patient groups.

Figure 3: Box-and-whisker plots of the mean parameters, D_m, α, β based on classical CTRW analysis in a) and the mean parameters, D_perf, f, D_m, α, β based on DISMANTLE in b) for the benign and malignant patient groups. The corresponding descriptive statistics, showing sample mean and standard deviation, (x±σ), of each parameter are summarized in the tables. The p-values for the statistical significance of the difference between the parameter values for the two patient groups are indicated in the last rows the tables; and the statistically significant p-values are highlighted in red.

Figure 4: The ROC curves of using the different combinations of classical CTRW-based parameters, D_m, α, and β, in a); classical IVIM-based parameters, D_diff, D_perf, f, in b); and DISMANTLE-based parameters, D_perf, f, D_m, α, and β in c) for differentiation between benign and malignant breast lesions. The corresponding tables summarize the sensitivity, specificity, accuracy, and AUC for the ROC curves. For each ROC analysis, the best performance metrics are highlighted in red.

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

0139