Diagnostic value of multiple b-value diffusion-weighted imaging in discriminating the malignant from benign breast lesions

The risk of cancer diagnosis has increased globally, and breast cancer remains the most frequently diagnosed female cancer that takes up almost 23% of the whole positive cases [1]. The breast cancer mortality rate shows a significant increase in 25 years throughout the world, increasing tendency in incidence and prevalence of this cancer account for the high mortality rate. Owing to its highly heterogeneous and metastatic character, it could commonly metastasize to distant organs, which was directly responsible for its malignance. Previous studies have reported that the 5-year survival rate of breast cancer patients is over 80% owing to the timely detection of this disease in North American [2]. Therefore, the early diagnosis in discriminating the benign breast lesions from malignancy was of considerable importance, which was helpful to its clinical treatment, good prognosis as well as increased survival rate.

Mammography was used as a convenient mean to screen the early-stage breast cancer. However, the dense breast tissue possibly obscured the potential mass in symptomatic women, resulting in the false-negative rate ranging from 8 to 66% [3, 4]. Some previous studies focused on the short acquisition time of ultrasound (US) imaging. However, it is seldom clinically used to evaluate the dynamic information of a contrast agent in lesions [5‐7]. Nevertheless, even though magnetic resonance imaging (MRI) or Dynamic Contrast Enhanced MRI (DCE-MRI) technique could generate superior results to analyze the breast lesions and compensate for its insufficiency of ultrasound examination, abundant in the spatial and temporal variation, the disadvantages of DCE-MRI could not be neglected [8]. The nephrogenic fibrosing dermopathy was correlated to the Gd-DTPA that has been extensively used worldwide in MRI evaluation as a component of intravenously administered contrast agents [9]. Some patients who received intravenous injection may show severe anaphylactoid reaction [10]. Moreover, the ongoing discussion about contrast agents intravenous exposure associated with neuronal tissue deposition even in the setting of relatively normal renal function and its deposition in the brain additionally highlights the need for alternative [11].

Nowadays, diffusion-weighted MRI (DWI) has been recognized as an attractive non-invasive, quantitative, adjuvant technique reflecting the functional information about the Brownian motion of water molecules as well as the microscopic organization and cellularity of biologic tissues [12]. Numerous evidence has found that DWI can address some limitations of conventional breast MRI by offering complementary information with short acquisition time and highly diagnostic sensitivity for lesions assessment. The ADC value would be used as an effective parameter to distinguish between malignant and benign breast lesions [13, 14]. Whereas the monoexponential model applied to conventional DWI was limited by a single b value and the threshold obtained to distinguish malignancy varied from person to person [15, 16], which led to the bias that the dispersive result could partly influenced by tissue perfusion. The emergence of the biexponential model-intravovel incoherent motion (IVIM) proposed by Le Bihan et al. contributed significantly to improving the effectiveness of microcirculation, the microscopic blood flow in tissue with rich perfusion [17]. IVIM technique can separate microvessel perfusion from diffusion by setting a range of b value. The quantitative parameters included microvascular volume fraction (f), molecular diffusion coefficient (ADC_slow) and perfusion-related incoherent microcirculation (ADC_fast). ADC_slow represents the mobility of water molecules in tissue and depends on the cellularity, tortuosity of the extracellular space, integrity of cell membranes, and viscosity of fluids, f reflects the relative contribution of microvascular blood flow to the DWI signal, and ADC_fast depends on blood velocity and length of microvessel segments [18]. To tackle the limitation of the hypothesis of two diffusion compartments [19], Bennett et al. proposed a stretched-exponential model as supplementary and it has been widely applied to the brain, liver and prostate but rarely in the breast [20, 21]. The distribution diffusion coefficient (DDC) and the heterogeneity index (α) were used to describe the behavior of signal attenuation analytically as a function of b values. The DDC derived from the fitting stretched-exponential function to the data was associated with ADC_slow [22]. Moreover, with the effect of ultra-high b-value, aquaporins (AQP) reflecting physiologically osmotic water transport across cell plasma membranes at cellular level, facilitating transepithelial fluid transport and its expression became more sensitive to the change of functional b values [23, 24]. Mostly research focused on the different subtypes of AQP expression in vivo [25], however, there was a lack of studies on ADC_AQP as a biomarker differently expressed for breast cancer prediction [26]. Therefore, various elements should be considered when distinguishing the doubtful breast lesion.

Our study aimed to evaluate whether multi-b-values of DWI parameters could provide more useful information to an optimal diagnostic fitting model capable of accurately differentiating the benign breast lesions from malignancy and in-depth explored multiparametric combination as a credible non-invasive examination was beneficial to decrease the unnecessary risk of biopsies.

A total of 79 lesions were confirmed by pathological examination clinically. The 38 patients (median, 48.5 years; range, 42.25–56.50 years) with breast cancer comprised 21 invasive ductal carcinoma (IDC) cases, 1 ductal carcinoma in suit (DCIS) case, 1 squamous carcinoma case and the remaining 15 cases were invasive ductal carcinomas with carcinoma in suit. While benign group (median 47 years; range, 38–50 years) classifications consisted of breast 20 fibroadenomas cases, 16 cystic hyperplasias cases, and 5 adenosis of mammary cases.

The comparison of clinical and conventional imaging characteristics between benign and malignant groups was shown on Table 1. It reveals that the clinical and some conventional imaging characteristics show no significance. The quantitative parameters derived from multiple b-value measurement data underwent Shapiro–Wilk test and satisfy normal distribution or nearly normal distribution (P > 0.05). The intraclass correlation coefficient (ICCs) of all parameters were raging from 0.801 to 0.964, which represented the measurements of multiparameter had a good interobserver reproducibility. Details were shown in Table 2. Table 3 showed the mean difference between malignant and benign groups. The ADC_avg (0.016 ± 0.0003 vs. 0.013 ± 0.0005), ADC_max (0.0019 ± 0.0003 vs. 0.0015 ± 0.0003) and α (0.7317 ± 0.1875 vs. 0.6497 ± 0.1767) values of benign lesion group were larger than those of the malignant lesions group (P < 0.05).While the ADC_fast value of the benign group was significantly lower than the malignant measurement (0.0133 ± 0.0158 vs. 0.0428 ± 0.0741) (P < 0.05). Conversely, the f value of benign tumor obviously larger (0.05940 ± 0.1410 vs. 0.4948 ± 0.1744) (P < 0.05). The ADC_AQP, ADC_min, ADC_slow and DDC values within two groups exhibited no statistical significances (P > 0.05). The box charts based on the analytical result above were portrayed in Fig. 1 and directly reflected the distribution of each parameter between benign and malignant lesions, offering valuable reference information.

The pathological results of breast lesions were recognized as the gold standard. Multiparametric combination of monoexponetial, biexponetial and stretched exponential parameters based on the statistical difference were summarized in Table 4. Interestingly, observing the combined AUC of the two parameters (ADC_avg-ADC_fast) (0.807; 95% confidence interval, 0.707–0.901) (P < 0.05) identified with the AUC of three parameters (ADC_avg-ADC_fast-α) (0.807; 95% confidence interval, 0.703–0.887) displayed the largest AUC and both of them were considerably greater than that of each parameter alone. Nevertheless, the Youden index of ADC_avg-ADC_fast was higher than that of ADC_avg-ADC_fast-α, maintaining the specificity of 87.8% and sensitivity of 68.42%. On the basis, this position would be determined as the cut-off point. Then, the AUC of ADC_avg (0.806, 95% confidence interval, 0.702–0.887) closely followed. The predictive values of the other multiparametric combinations or independent parameter, with AUCs below 0.80. Besides, ADC_avg-max-α was the best specificity with the highest average of 92.68% but lowest sensitivity of 55.26% on the contrary. The ROC curves of the mono-exponential, bi-exponential and stretched-exponential parameters were drawn in Fig. 2, all of which was selected with good diagnostic performance (AUC at least for 0.75) and compared with parameter involved alone. The result ultimately revealed that the diagnostic efficacy of combination of ADC_avg-ADC_fast, ADC_max-α, ADC_max-ADC_fast-f were markedly greater (P < 0.05). Figures 3 and 4 provided functional parameter maps of benign and malignant breast lesions.

Diffusion-weighted MRI has currently been incorporated into breast MRI protocols to improve the sensitivity of potential cancer whose ADC value has been considered as a quantitative index to address some deficiencies of routine clinical breast MRI. Compared with other detective techniques, the non-contrast and predictive assessment of differentiation of benign and malignant breast lesion showed superiority [13]. Mebis et al. [28] explored the correlation between average and minimum ADC values on MRI and they both showed significant differences in disease classification. To investigate the possible influence on microcirculation to molecular diffusion motion, we took ADC_avg, ADC_min and ADC_max value of monoexponential diffusion model into consideration. The result demonstrated that it was the ADC_avg value that had an AUC of 0.806 and the better Youden index with a sensitivity of 57.89% and specificity of 97.56%, indicating that ADC_avg value was a reliable quantitative measurement for predicting malignant breast lesions preoperatively. Nonetheless, it deviated from the previous research conducted by Rupa R et al. who found that the minimum ADC value was reported to be more accurate in classifying the grade of breast tumor than mean ADC value [29]. The variation revealed that clear consensus had yet been reached regarding this association, suggesting that the perfusion was somehow likely to have impact on the ADC value.

Multiparametric MRI diffusion models have been currently served as a feasible technique to characterize quantitatively of lesions, such as liver fibrosis, prostate cancer, and salivary gland lesions [30‐32]. The ADC_slow, ADC_fast value and f derived from biexponential model represented tissue true diffusivity, perfusion-related incoherent microcirculation along with perfusion fraction respectively while even though ADC_slow value (0.0012 ± 0.0014 vs. 0.0008 ± 0.0007, P = 0.1830) of IVIM manifested increasing cellularity and confined extracellular spaces of breast cancer, it showed no statistical difference between benign and malignant lesions, which was inconsistent with the previous study [32]. Perhaps the rough definition failed to sort out the mean or 50th percentile of ADC_slow and certain overlap existed between groups existed, deprived of reflecting the realistic heterogeneity [33].

Moreover, it was because we separated the diffusion from perfusion through high b-values drastically that ADC_fast value (0.0428 ± 0.0741, P < 0.015) of malignant breast carcinoma was higher than benign tumor in our study. Meanwhile, most studies supported that abundant blood capillaries in tumor tissues might give a probable explanation that the high perfusion replenished the constrained liquid movement [34]. However, the f value (0.5940 ± 0.1401 vs. 0.4948 ± 0.1744, P<0.0060) of benign lesions was higher than malignancy, which was in contrast to the conventional assumption [35]. The possible reason was that on the one hand, the vessels from malignant tumors were oriented to small, leaky and few efficient tumor capillaries and tortuous vascular hyperplasia along with farraginous vascular branches in malignant lesion contributed to it [36]. On the other hand, it was indicated that the microvascular compression derived from the multitudinous cell density resulted in the lower f value [37]. The short TE acquisitions (TE<100 msec) in our study compared to the TE = 103 msec used by Sigmund et al. [38], as shown by Lemke et al. [39], might also result in the variability of low f value. Furthermore, Bokacheva L et al. had implied that lesions ROIs excluding the tumor edges conduced to lower perfusion fraction due to the tumor periphery were more vascularized than the red tumor center [40]. Some scholars also inferred that the change of ADC_fast was dependent on f, probably on account of the f signifying the blood-carrying capacity of the capillaries, but the ADC_fast reflecting the flow rate of blood, which changed when the microvascular diameter changed with volume invariability. Interestingly it could explain our data observed at length [41, 42].

We further reported the stretched exponential model that characterized the non-Gaussian behavior of molecular diffusion and reflected the degree of intravoxel heterogeneity of biological tissue for breast cancer. Bennnett et al. put forward a hypothesis that closer α to 1 and the higher homogeneity it would be [22]. In this study, α values of malignant lesions were lower than those of benign structures, for the sake of the fact that malignant lesions were associated with considerable histological heterogeneity. The high degree of cellular pleomorphism indicated high variability, the existence of intravoxel microscopic cystic or necrotic foci and even the heterogeneity in vascular structures [43, 44]. Besides, the DDC values were statistically higher in the malignant group while α exhibited lower in malignancy as compared with the benign group, which disagreed with the adopted evidence [45, 46]. The result suggested the inverse correlation between tumor cellularity and diffusion coefficient, indicating that the inhomogeneous necrosis and cystic components emerged in the target area. Thus, the negative relationship warranted further lager cohort validation, and histopathological correlation would comprehend the puzzle to some extent.

Additionally, AQP involved in cell migration for tumor angiogenesis and local invasion took charge of the water transport through membranes. When ultra-high b-values offered, the expression of AQP must be affected [26]. The ADC_AQP values (0.3636 ± 0.2175 vs 0.3699 ± 0.1139) of breast cancer were higher than the benign group, which was in support of the idea that AQP facilitated cell migration, not only relevant to angiogenesis but also to tumor spread, glial scarring, would healing and any other phenomena containing immune-cell chemotaxis [47]. Whereas the ADC_AQP showed no statistical differences in the current test. It would depend upon that the quantitative diffusion fitting model might be influenced by both target tissue and the choice of optimal b-value. Moreover, the inclusion of high b-values in our study prolonged the TE generating the consequent decrease in SNR. Previously published studies for preferable b value of breast lesion were still insufficient for all diffusion model, which needed to be optimized through in-depth research [46].

We considered that the comparison among the multiparametric utilization should establish criteria presenting a good differential diagnostic ability with not only the AUC greater than 0.75 but also outperforming any other quantitative parameters used in isolation. The required ROC analysis demonstrated that the ADC_avg still achieved good diagnostic efficacy (AUC = 0.806) with the highest specificity (97.56%) for breast cancer detections ranking only second to the multiparametric combination (ADC_avg-ADC_fast, AUC = 0.807). However, the excellent specificity was at the expense of decreased sensitivity for the sake of the reduction of false positive rate. Fornasa et al. [48] had reported fat necrosis exhibited malignant features like irregular mass and architectural distortion in accordance with our result. With regard to the combination of multiple variables, they might lead to a higher diagnostic power compared to the utilization of single parametric features. Surprisingly, the diagnostic efficacy of ADC_avg-ADC_fast was the same as ADC_avg-ADC_fast-α (AUC = 0.807) in our study, signifying that the simplified model could also warrant the heterogeneity of malignant lesions. Furthermore, compared with other routine quantitative measurements, the highest Youden indices of ADC_avg-ADC_fast with a sensitivity of 68.42% and specificity of 87.8% implied the superiority of differential diagnosis available in clinical practice, increasing its feasibility as a non-invasive tool.

There were several limitations to the current study. First, our investigation was constrained by relatively small sample size. The prospective validation of multiparametric combination model needed an overwhelming amount of data and develop a nomogram as supplement to present variables visually. Second, the stage of breast lesions was not respectively evaluated, so whether the lymph node metastases had existed was likely to become an influencing factor as Kamitani et al. [49] had observed that diffusion coefficients were higher in cases that were positive for axillary lymph node, and where micro-necrosis and fibrosis inside the lesion mattered. Third, neither age nor menstrual cycle were controlled in the enrollment. Forth, this developed model needed a further validation. Lastly, the choice of b-values was not optimized for all diffusion models. Generally, the b-values in our study were used to meet a wide range for diffusion. This study suggested that a comprehensive evaluation of breast cancer patients using advanced imaging may increase insight into tumor physiology. It was important to make the individualized treatment plans for high-risk women and such finding paved the way for deeper investigation of the potential of multiparametric MRI in the differentiation of breast cancer.

In conclusion, our research proved that multiple b-values diffusional exponential model contributed to the differential diagnosis of malignant breast lesions and multiparametric quantitative imaging with the combination of ADC_avg and ADC _fast could enhance the diagnostic ability of breast cancer detection rate by reflecting more biological characteristics of breast tissue and lesions. Therefore, if multiparametric technique combined with the utilization of ADC_avg-ADC_fast was implemented into standard scanning protocol may have opportunity to maximize diagnostic accuracy while avoiding unnecessary breast biopsies.