Characterization of Breast Lesions Using Diffusion Kurtosis Model-Based Imaging

Methods: This prospective study included 120 consecutive patients underwent preoperative DCE-MRI examinations and multi-b-value diffusion-weighted imaging (DWI). 88 maligant lesions and 44 benign lesions were found and 56 normal fibroglandular breast tissue were selected as normal control. Conventional apparent diffusion coefficient (ADC) as well as DKI-based parameters mean kurtosis (MK) and mean diffusivity (MD) were analyzed by lesions types and histological subtypes using one-way ANOVA and receiver operating characteristic (ROC) curve.


Introduction
Breast cancer has become the most common form of tumors in women [1], and Dynamic contrast enhanced (DCE) magnetic resonance imaging (MRI) is an rather sensitive technique method for early detection and identification of benign and malignant breast lesions. However, diagnostic difficulty may exist in identification of benign and malignant breast lesions due to the overlap of time intensity curves of benign and malignant lesions.
MR diffusion-weighted imaging (DWI) is of value in diagnosing of benign and malignant breast lesions, and its apparent diffusion coefficient (ADC) can improve the specificity of dynamic contrastenhanced (DCE) MRI in the diagnosis of malignant breast lesions [2][3][4]. However, there is also a substantial overlap in ADC values for benign and malignant lesions. The traditional DWI is based on the assumption that the diffusion of water molecules (random Brownian motion) within tissues is in accordance with a Gaussian distribution, whereas restrictions imposed by cell membranes, organelles, shrunked intercellular spaces due to cell proliferation and complex structures in tumor tissues result in the diffusion of water molecules deviating from a Gaussian distribution.
Acknowledging this problem, Jensen et al. [5,6] introduced a diffusion kurtosis imaging (DKI) model which is an extension of Difussion Tenser Imaging (DTI) to quantitatively describe the displacement of water molecular diffusion that deviated from a Gaussian distribution. Mean kurtosis and mean diffusion coeffcients are calculated from this model.
Mean Kurtosis, which is a dimensionless metric, quantifies the deviation of water molecular diffusion from a Gaussian distribution and can potentially be sensitive to some tissue properties, such as heterogeneity [6]. Mean diffusivity is the diffusion co-effcient with correction of non-Gaussian bias. Researches have been conducted with DKI model on different organs such as brain [7][8][9][10], neck [11,12], kidney [13,14], prostate [15,16], and breast [17,18]. These studies showed DKI parameters are able to improve performance of differentiating malignant from benign lesions than traditional DWI and to allow differentiation among tumor grades such as cerebral glioma. Given that DKI provides information about diffusion that is closer to the actual state, its clinical utility might be superior to that of traditional DWI. However, studies of the accuracy of this model and the correlation between the DKI-derived parameters and different pathologic subtypes are lacking. Therefore, this present prospective study was designed to investigate the ability of DKI to differentiate between benign and malignant breast lesions, and to assess different pathological subtypes of breast cancer and different grades of invasive ductal carcinoma (IDC) [19].

MRI Examination
MRI examinations were performed in a prone position on a 1.5T MRI system (Aera, Siemens, Germany) with a dedicated fourchannel bilateral breast coil in the axial orientation. The scanning protocol was as follows:

i.
A routine scan was performed using a fat-suppressed

Pathology
In order to ensure that the area measured in the MRI images was consistent with the area examined postoperatively in pathological sections, the upper, lower and inner edges of the lesion were labeled during surgery, and the tumor was completely resected.
According to the lesion location in the body, the lesion specimens in vitro were stereo-positioned based on the intraoperative labels, and transverse pathological sections were obtained. The size, pathological type and histological grade of the tumor were recorded.
Semi-quantitative evaluation and classification were performed using the histological grading system described by Elston and Ellis [6]. Glandular formation, nuclear polymorphism and mitotic count were each scored I to III, and the grade was determined from the overall score: grade I, 3-5; grade II, 6-7; and grade III, 8-9.

Comparison of Each Parameter Between Normal, Benign and Malignant Groups
There were significant differences in ADC, MK and MD values between normal, benign and malignant groups (P < 0.001). In pairwise comparisons, ADC and MD were significantly lower in the malignant group than in the benign and normal groups, whereas MK was significantly higher in the malignant group than in the benign and normal groups (Table 1 and

Diagnosis of Malignant Breast Lesions
The AUC for diagnosing malignant breast lesions was 0.936 for MD, 0.911 for MK and 0.897 for ADC ( Figure 6); the only significant difference was for MD vs. ADC (P = 0.015). The sensitivity, specificity, positive predictive value, negative predictive value and accuracy of each parameter for the diagnosis of malignant breast lesions were shown in Table 2.      h) Pathological image routinely stained with hematoxylin-eosin (×100) showing invasive growth of tumor tissue, a large mucous lake, a floating heterotypic cell mass on the mucous lake, adenoid-like structure in some regions, nuclear hyperchromatism and atypical heterocysts, and surrounding fibrous tissue hyperplasia.

Discussion
The malignant lesions possess lower MD value and higher MK value than benign lesions and normal tissues, which are consistent with the results of Nogueira et al. and Sun et al. [18,21] We found that the diagnostic efficacy for malignant breast lesions was higher for DKI model than for ADC and MK, which showed higher sensitivity and MD showed higher specificity.
However, our results were different from those of Wu et al. [24].
Conventional mono-exponential model neglects heterogeneous diffusion distribution of water molecules in tissue resulting in great overlap in ADC values for benign and malignant lesions.
However, the DKI model provides a quantitative description of the extent to which water molecule diffusion deviates from a Gaussian distribution, hence may reflect the complexity of the voxel microstructure, which is a more realistic reflection of the state of the body. This might explain why DKI model is superior to conventional DWI in diagnostic efficacy. The MD value was higher than ADC value in our study and sensitivity and specificity of MD were higher than ADC, which suggests that some important information may be implied by MD. Wu et al. [24] reported that the sensitivity and specificity of MK for distinguishing between benign and malignant breast lesions was higher than that of MD, and the misdiagnosis rate was lower. Nevertheless, the results of our study showed that the AUC of MD (0.936) in the diagnosis of malignant breast lesions was numerically (but not significantly) higher than that of MK (0.911) and significantly higher than that of ADC (0.897). The specificity of MD was very high (98.3%), and the sensitivity, positive predictive value, negative predictive value and accuracy of MD were higher than those of ADC. These results suggest that DKI may provide more valuable information than conventional DWI with regard to microstructural changes. Thus, both MK and MD values may have important clinical utility in the differential diagnosis of benign and malignant breast lesions. The MK value had higher sensitivity, while the MD value had higher specificity, which would improve the differentiation of these lesions.
The findings of our study demonstrated that the MD value increased successively from IDC to DCIS to MBC. Adversely, the MK value successively decreased. The differences between these pathological subtypes were significant. Previous studies showed that ADC value increased in turn from IDC to DCIS to MBC [25][26][27][28], which MD value in our study is consistent with that. The carcinomatous ductal epithelium of DCIS is restricted in situ and does not break through the basement membrane, whereas IDC disrupts the basement membrane through proteolytic activity and extensively invades and spreads across the mesenchyme. On the other hand, proteolysis-induced chronic inflammatory infiltration promotes the proliferation of connective tissue and leads to a further decrease in the extracellular space and a higher density of cells [28]. As a result, ADC and MD are lower for IDC than for DCIS.
The MK value for IDC is higher than DCIS. It suggests that DCIS and IDC could be identified preoperative according to the MD and MK values and provides clinician with decision reference to avoid overtreatment. MBC tumors contain a large amount of extracellular mucus components with a very low cell density [27,29] and have a simple tissue structure with low restriction of water molecule diffusion. Previous investigations have demonstrated that the ADC value was higher for MBC than for benign tumors and other malignant tumors [27]. This study showed that the performance characteristics of MD, an averaged diffusion co-efficient in each direction after calibration for a non-Gaussian distribution, were similar to those of ADC. The MD value of MBC was higher than that of IDC, DCIS and benign lesions, while the MK value was lower. SPC is an intraductal lesion, and there were no significant differences in all parameters between SPC and DCIS. SPC is characterized by a compact arrangement and expansive growth, is rich in cellular nodules, has a fibrovessel axis, and is often accompanied by mucinous carcinomas and/or neuroendocrine carcinomas [30,31], making it heterogeneous. As a result, it is difficult to distinguish between SPC and IDC. Our study found that SPC and IDC showed no differences in ADC and MK but a significant difference in MD. This suggests that the MD value, which is based on the complexity of the biological tissue microstructure, may provide additional valuable information for breast lesions with similar ADC values.
This study revealed that there was no correlation between DKI-model parameters and histological grade of IDC, which is inconsistent with the results of Sun et al. [21]. Sun reported that the MK and MD values of grade III IDC were significantly different from those of grades I and II. Sun also suggested that the MK value was positively related to histological grade, and MD was negatively related to the histological grade of IDC. Although no cases of grade I in this group of randomly selected cases may result in analysis bias, grades I and II were integrated into lower grade group in Sun's analysis. Thus, the results should be interpreted with caution.
Studies [32,33] have shown that attenuation of the MR signal in highly perfused tissues results from the combined effects of water molecule diffusion and microvascular perfusion. Attenuation of the DW signal by microvascular perfusion is due to spin dephasing induced by false blood diffusion within the blood capillary of the voxel, with intravoxel incoherent motion. However, blood perfusion mainly exhibits false diffusion at low b-values, and because of this, higher b-values (600, 1200, 1800 and 2400 s/mm 2 ) were used in this study to fit the attenuation of the DW signals for kurtosis imaging analysis so as to minimize the spurious diffusional effect of blood perfusion. Moreover, the scan time was only 70s, which is acceptable for most patients. Therefore, it would be feasible to apply this model of DKI to clinical practice.
The mammary glands gradually atrophy is replaced by adipose tissue with increasing age. Adipose tissues in mammary glands can produce a significantly lower diffusion co-efficient [34], and a poor effect of fat suppression might lead to overestimation of the MK value. Thus, Multi-b value DWI with STIR was performed in this study in order to acquire better fat-suppression uniformity.
This study has some limitations which need to be pointed out. First, this study did not evaluate the effects of menopausal status and the menstrual cycle on the utility of DKI parameters for the assessment of breast tumor tissues. It has been reported that the ADC value of a tumor does not vary with different stages of the menstrual cycle but is significantly lower in postmenopausal women than in premenopausal women [35]. Second, different mammary gland types may have an influence on the DKI parameters, and the MD value of heterogeneous dense glands may be higher than that of glands containing more adipose tissue. In this study, the influence of gland type on DKI parameters in the normal breast group was also not evaluated.
In conclusion, the diagnostic efficacies of the DKI parameters, MK and MD, were higher than that of ADC obtained with a conventional mono-exponential model, hence the use of MK and MD could improve differentiation between malignant and benign breast lesions. MD and MK could also potentially be used in the differential diagnosis of IDC, DCIS, MBC, IDC and SPC.