Histogram-metrical DCE-MRI based Quantitative Discrimination of Breast Masses

Chao Jin^{1}, Ting Liang^{1}, Hongwen Du^{1}, Gang Niu^{1}, Zihua Su^{1}, Peng Cao^{1}, Yonghao Du^{1}, Chenxia Li^{1}, Yitong Bian^{1}, and Jian Yang^{1}

^{}In this
study, a mean and 2 histogram metrics (median and variance) of K^{trans}, mean and all histogram metrics (skewness, kurtosis, median, entropy and energy)
of k_{ep} showed significantly differences between benign and malignant
masses (*p*_{max}=0.036) (**Figure 2**). With respect to IDC grading, significant differences were only found in histogram entropy and energy of kep between grade II- and III-IDC groups (entropy: median 4.38, range 3.82~4.66 v.s. median 4.54, range 4.33~4.69, *p*=0.0054; energy: median 0.015, range 0.011~0.03 v.s. median 0.012, range 0.01~0.016, *p*=0.011) (**Figure 2**).

In discrimination of benignancy and malignancy, the area under the ROC for mean of K^{trans} was 0.73, with a best 2D mean cutoff value of 0.192 that resulted in sensitivity of 79.17% and specificity of 66.67%. While the maximal area under the ROCs for histogram metrics of K^{trans} was 0.71 with a best cutoff value of 0.171 that resulted in sensitivity of 75.00% and specificity of 66.67% (**Table 1**). In terms of kep, the maximal areas under the ROCs for mean and histogram metrics were 0.83 and 0.91, respectively. Notably, two best cutoff values (i.e. entropy, 4.189; energy, 0.017) that yielded a higher sensitivity of 87.5% and specificity of 93.33% (**Table 1**). There were no significant differences between the areas under K^{trans} ROCs of mean and histogram metrics (*p*_{min}=0.576); or the k_{ep}’s mean and histogram metrics (*p*_{min}=0.157). While significant differences were observed between maximal AUCs of Ktrans and kep ( *p*=0.011).

In respect of IDC grading, the areas under kep ROCs of histogram entropy and energy were respectively 0.84 and 0.81 (entropy: sensitivity 70%, specificity 92.86%; energy, sensitivity 80%, specificity 78.57%) ( **Table 1**). No significant difference was also found between the areas under kep ROCs of such two metrics (*p*=0.210).

1. Berg WA, et al. Diagnostic accuracy of mammography, clinical examination, US, and MR imaging in preoperative assessment of breast cancer. Radiology. 2004;233:830–849.

2. Just N. Histogram analysis of the microvasculature of intracerebral human and murine glioma xenografts. Magn Reson Med, 2011, 65(3):778–789.

3. O'Connor J P B, et al. DCE-MRI biomarkers in the clinical evaluation of antiangiogenic and vascular disrupting agents. Brit J Cancer, 2007, 96(2): 189–195.

4. Li J, et al. A clinically feasible method to estimate pharmacokinetic parameters in breast cancer. Med Phys. 2009;36:3786–3794.

5. Tofts P S, Berkowitz B, Schnall M D. Quantitative analysis of dynamic Gd-DTPA enhancement in breast tumors using a permeability model. Magn Reson Med, 1995, 33(4): 564-568.

Figure 1. Histogram
analysis of (A) benign tumor (female, 41 yr, breast fibroma) and (B) malignant
tumor (female, 53 yr, grade III-IDC). For each, on the top are DCE MR image,
K^{trans} and k_{ep} maps from left to right; on the bottom are histogram
contributions of K^{trans} and k_{ep} values.

Figure 2. *Z*-scores
of mean and histogram metrics of (A) K^{trans} and (B) k_{ep} in discrimination of
benignancy and malignancy; (C) *z*-scores of k_{ep} mean and histogram metrics in
discrimination of grade II- and III-IDCs. * *p*<0.05.

Table 1. Comparison of areas under the ROCs of mean and histogram metrics: benignancy
v.s. maglinancy; grade II-IDC v.s. grade III-IDC

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

2731