Diffusional kurtosis imaging for differentiation of additional suspicious lesions on preoperative breast MRI of patients with known breast cancer

Abstract

Purpose:

To investigate the potential of diffusional kurtosis imaging (DKI) and conventional diffusion-weighted imaging (DWI) in the evaluation of additional suspicious lesions at preoperative breast magnetic resonance imaging (MRI) in patients with breast cancer.

Materials and methods:

Fifty-three additional suspicious lesions in 45 patients with breast cancer, which were detected on preoperative breast MRI, were examined with a 3-T MR system. DKI and DWI data were obtained using a spin-echo single-shot echo-planar imaging sequence with b-values of 0, 50, 600, 1000, and 3000 s/mm². Histogram parameters (mean, standard deviation, minimum, maximum, 10th, 25th, 50th, 75th, 90th percentiles, kurtosis, skewness and entropy) of ADC from DWI and diffusivity (D), kurtosis (K) from DKI were calculated after postprocessing. Parameters were compared between benign vs. ductal carcinoma in situ (DCIS) vs. invasive breast lesions and diagnostic performances were evaluated by receiver operating characteristic (ROC) analysis. Correlation between the mean values of D and K was analyzed according to lesion type.

Results:

Multiple histogram parameters of D (mean, 25th, 50th percentile, 75th percentile, and entropy) differed between benign and invasive breast lesions (all P < 0.005), but none differed between benign vs. DCIS. D-90th percentile differed between DCIS vs. invasive cancer (P = 0.040). K-10th percentile differed between benign vs. DCIS (P = 0.015). ADC-75th percentile differed between benign vs. invasive cancer and ADC-75th percentile, ADC-90th percentile differed between DCIS vs. invasive cancer, respectively (all P < 0.005). ROC curve analysis showed high specificity for discrimination between benign and invasive cancer. D-mean and K-mean showed strong correlation in benign (r_s = −0.813) and invasive lesions (r_s = −0.853), but no significant correlation in DCIS.

Conclusion:

DKI may aid in the differentiation of additional suspicious lesions at preoperative breast MRI. Both ADC and DKI may have lower potential in differentiating DCIS from benign lesions.

1. Introduction

Breast magnetic resonance (MR) imaging is the most sensitive imaging tool for detecting breast cancer, with a sensitivity of close to 100% [1,2]. In patients with newly diagnosed breast cancer, dynamic contrast material-enhanced (DCE) MR imaging has been reported to detect additional disease in the ipsilateral breast in 6%–34% and in the contralateral breast in 3%–6% of patients [3–6]. Although the clinical importance of additional diseases detected at MR imaging is still under debate [7–11], the majority of multicentric cancers detected only on MR imaging are invasive cancers and approximately 25% have been reported to be larger than in 1 cm [12]. In a meta-analysis of 3253 women, 64.9% of MR imaging-detected contralateral cancers were also reported to be invasive cancers [5]. Yet, DCE-MR imaging has lower and variable specificity for diagnosing breast cancer, and false-positive findings may lead to additional investigations or surgery [5,6,13].

Diffusion-weighted imaging (DWI) has been proposed to improve the specificity of breast MRI exams [14,15]. Conventional DWI can be used to measure apparent diffusion coefficient (ADC) values, which have been shown to differ significantly between benign and malignant breast lesions [16]. Although conventional DWI assumes a Gaussian diffusion of water protons, water diffusion in complex biological tissues shows a non-Gaussian phenomena, likely associated with tissue microstructure [17,18]. Diffusion kurtosis imaging (DKI) is a non-Gaussian diffusion weighted analysis method and includes calculation of diffusivity (D, diffusion coefficient with correction of non-Gaussian bias) and kurtosis (K, deviation of tissue diffusion from a Gaussian pattern) [19]. Several recent studies have reported that DKI improved the characterization of breast lesions [18–20], with one study reporting a higher specificity of DKI than that of conventional DWI [19]. However, no previous studies have investigated whether DKI could also be used to differentiate additional suspicious lesions detected at preoperative breast DCE-MR imaging in breast cancer patients. Therefore, the purpose of this study was to investigate the potential of DKI and conventional DWI in the evaluation of additional suspicious lesions detected at preoperative breast MR imaging in patients with newly diagnosed breast cancer.

2. Materials and methods

2.1. Study population

This retrospective study was approved by the institutional review board, and the requirement for informed consent was waived. Between July 2014 and September 2016, 1096 consecutive women with newly diagnosed breast cancer underwent preoperative breast MR imaging using a 3-T MR scanner (Ingenia, Philips Medical Systems, Best, The Netherlands). Among them, 194 patients underwent biopsy or surgery for additional suspicious breast lesions detected by breast MR imaging. Considering challenges in depiction of small lesions due to limited spatial resolution and lower SNR with the use of higher b values [21–23], we excluded 149 patients with lesions smaller than 10 mm. Finally, 45 patients (mean age, 46 years; range, 29–65 years) with 53 pathologically confirmed (confirmed by surgery [n = 28], US-guided vacuum-assisted biopsy [n = 2] and core needle biopsy [n = 23]) additional suspicious breast lesions ≥10 mm, which were detected by preoperative breast MRI imaging, were included in our study.

2.2. MRI acquisition protocol

Imaging was performed by using a 3-T MR scanner (Ingenia, Philips Medical Systems, Best, The Netherlands) with a dedicated 16-channel bilateral breast coil with the patient in the prone position. Sequences included a three-plane localizing sequence, axial bilateral modified Dixon turbo spin echo T2-weighted sequences and axial T1-weighted fat-suppressed dynamic contrast-enhanced sequence with one pre-contrast and six post-contrast acquisitions. A diffusion MRI study was performed before dynamic contrast enhanced MRI, using a single-shot spin-echo EPI pulse sequence with the scan parameters shown in Table 1. Sensitizing diffusion gradients were applied in three orthogonal directions and trace-weighted diffusion-weighted images were generated.

Table 1.

Diffusion weighted imaging (DWI) sequence used for conventional DWI and diffusion kurtosis imaging.

Parameter	Diffusion sequence
Sequence	Single-shot spin echo EPI
Orientation	Axial bilateral
TR/TE (ms)	14275/121
Fat suppression	SPAIR
Field of view (mm2)	320 × 320
Matrix	224 × 227
Slice thickness (mm)	3
Number of signals averaged	1
Number of slices	50
Bandwidth (Hz/pixel)	1152
Scan time (min)	7:07
b-values (sec/mm2)	0, 50, 600, 1000, 3000
Acceleration factor	2
Parallel imaging technique	Sensitivity Encoding (SENSE)

2.3. Image analysis

For a conventional DWI measure, ADC maps were generated with a set of diffusion weighting factors (b values) of 50, 600, and 1000 s/mm². For DKI, diffusivity (D) and kurtosis (K) maps were calculated with b values of 50, 600, 1000, 3000 s/mm². To obtain the DKI parameters, a voxel-by-voxel fit was performed with the signal decay modeled by the cumulant expansion of the signal [24]:

$$\text{In}\left[S\left(b\right)\right]=\text{In}\left[S\left(0\right)\right]-bD+\frac{1}{6}{b}^2{D}^{2}K+\cdots$$

where S(b) is DWI signal at a particular b value, S(0) the baseline signal without diffusion weighting, D diffusivity, and K diffusional kurtosis. D represents the diffusion coefficient with correction of non-Gaussian bias. K, the first higher-order term in the cumulant expansion, is a dimensionless parameter that quantifies deviation of water motion from a Gaussian distribution. When K is equal to zero, mean D becomes equal to ADC and indicates a perfect Gaussian distribution. A larger K value indicates a larger deviation of diffusion from a Gaussian pattern. ADC, D and K were estimated using the weighted linear least square method using an in-house developed software program in MATLAB (The MathWorks, Natick, MA) [25]. Prior to estimating D and K, ADC was estimated first for all voxels. Any voxel with ADC > 3.5 (higher than free water diffusivity) was not included in further analysis for D and K estimation, since such high ADC values are not physically possible and must be due to noise or any other artifact.

DWI data were analyzed by two radiologists (M.J.K and V.Y. P, with 16 years and 4 years of experience in breast imaging, respectively) to identify all lesions in consensus by reviewing the DWI images, with reference to the contrast-enhanced T1-weighted images. Regions of interest (ROI) for each lesion were manually drawn on a representative slice on DWI raw images and were copied onto diffusivity (D) maps and kurtosis (K) maps. The mean ROI size for additional suspicious lesions on preoperative breast MRI was 77.8 mm² (range, 18.1–447 mm²). We also drew an ROI for each index breast cancer lesion, except for three cases in which the primary lesion had underwent surgical excision (n = 1) or vacuum-assisted biopsy (n = 2). The mean ROI size of index breast cancer lesions was 166.5 mm² (range, 20.3–796.5 mm²).

2.4. Statistical analysis

Histogram analysis was applied to ADC, D, and K. The histogram measures included in this study were mean, standard deviation (SD), minimum, maximum, percentiles (10th, 25th, 50th, 75th, and 90th), kurtosis, skewness and entropy. We compared the histogram measures of ADC, D and K between benign vs. ductal carcinoma in situ (DCIS) vs. invasive breast lesions among additional suspicious lesions by using a nonparametric multiple comparison test (Kruskal-Wallis test followed by the Dunn multiple comparison test). For the parameters that showed a significant difference, we performed a receiver operating characteristic (ROC) curve analysis. Sensitivity and specificity were calculated with a threshold criterion determined by using the maximum Youden index. Correlation between the mean values of D and K was analyzed by Spearman coefficient (rs) according to lesion type. In addition, for 40 additional suspicious lesions which were detected in patients with invasive index breast cancer, correlations of the mean values of ADC, D and K between the index cancer and additional suspicious lesions were analyzed using the Spearman coefficient (rs) according to lesion type.

A two-tailed P-value of < 0.05 was considered to indicate a statistically significant difference. Statistical analyses were conducted using SPSS for Windows, version 23.0 (IBM corporation, Armonk, NY) and MedCalc Statistical Software version 18.2.1 (MedCalc Software bvba, Ostend, Belgium).

3. Results

3.1. Lesion characteristics

Of the 53 pathologically confirmed additional suspicious lesions detected on preoperative breast MR imaging, 23 (43.4%) were benign and 30 (56.6%) lesions were malignant. The benign lesions consisted of fibroadenomas (n = 7), fibroadenomatous hyperplasia (n = 7), fibrocystic disease (n = 3), sclerosing adenosis (n = 1), intraductal papillomas (n = 2), radial scar (n = 1), atypical ductal hyperplasia (n = 1) and lobular carcinoma in situ (n = 1). The malignant lesions included ductal carcinoma in situ (n = 14, 46.7%) and invasive breast cancer (n = 16, 53.3%) [invasive ductal carcinoma (n = 11), invasive lobular carcinoma (n = 3) and mucinous carcinoma (n = 2)] (Figs. 1, 2). All of the malignant lesions and 5 (21.7%, including lobular carcinoma in situ [n = 1] and atypical ductal hyperplasia [n = 1]) of the benign lesions underwent surgical excision. The median follow-up period for benign breast lesions was 25.0 months (range, 5.8–40.1 months). The median lesion size on MRI was 12 mm (range, 10–92 mm).

Fig. 1.

Images in a 45-year-old woman with an additional MR-detected suspicious lesion in the left subareolar breast. (a) Axial T1-weighted contrast-enhanced subtracted MR images show the main 32-mm invasive breast cancer at the left outer central breast (posterior arrow) and a 15-mm suspicious lesion (anterior arrow) in the left subareolar breast, which was additionally detected on preoperative breast MR imaging. DWI image with b-values of 1000 s/mm² (b) and 3000 s/mm² (c) show the left subareolar breast lesion (arrow) as hyperintense. ADC (d) and diffusivity (e) maps, respectively, show decreased signal intensity of the left subareolar breast lesion (arrow) compared with surrounding glandular tissue. (f) Kurtosis map shows that the lesion is slightly hyperintense. The lesion was surgically confirmed as ductal carcinoma in situ.

Fig. 2.

Images in a 40-year-old woman with newly diagnosed breast cancer who underwent preoperative MR imaging. (a) Axial T1-weighted contrast-enhanced subtracted MR image shows an additionally detected lesion in the right lower central breast (arrow). DWI images with b-values of 1000 s/mm² (b) and 3000 s/mm²(c) show the right lower central breast lesion (arrow) is hyperintense. ADC (d), diffusivity (e) and kurtosis maps (f), respectively, show the right lower central breast lesion. The lesion was surgically confirmed as an intraductal papilloma with infarcted feature.

Of the 45 index breast cancer lesions, 35 lesions were invasive breast cancer (invasive ductal carcinoma [n = 30], invasive lobular carcinoma [n = 2], mucinous carcinoma [n = 2], invasive micro-papillary carcinoma [n = 1]) and 10 lesions were ductal carcinoma in situ. Among the 10 index DCIS lesions, three had underwent surgical excision (n = 1) or vacuum-assisted biopsy (n = 2), and were therefore excluded from ROI and statistical analysis.

3.2. Comparison of ADC and DKI: Benign vs. DCIS vs. invasive breast cancer

When comparing ADC histogram parameters among benign vs. DCIS vs. invasive breast cancer, the Kruskal-Wallis test showed significant differences in the mean (P = 0.028), 50th percentile (P = 0.048), 75th percentile (P = 0.014) and 90th percentile (P = 0.037) of ADC between them (Table 2). Results of multiple comparison testing showed that the mean ADC in invasive cancer tended to be lower than that of benign breast lesions (P = 0.059) and DCIS (P = 0.059). The 50th percentile of ADC in invasive breast cancer tended to be lower than that of benign lesions (P = 0.059). The 75th percentile of ADC in invasive breast cancer was significantly lower than that of both benign breast lesions (P = 0.027) and DCIS (P = 0.041). The 90th percentile of ADC in invasive breast cancer was significantly lower than that of DCIS (P = 0.038), but did not significantly differ with that of benign lesions. None of the ADC histogram parameters showed a significant difference between benign lesions and DCIS.

Table 2.

Comparison of ADC histogram parameters between benign vs. DCIS vs. invasive breast cancer.

	Benign (n = 23)	DCIS (n = 14)	Invasive (n = 16)	P value	P valuea
					Benign vs. DCIS	Benign vs. Invasive	DCIS vs. Invasive
Size (mm)	12.0 (10–50)	13.5 (10–92)	12.0 (10–30)	0.458
ADC parameter (10−3mm2/s)
Mean	1.04 (0.68–2.09)	1.10 (0.79–1.33)	0.83 (0.52–2.30)	0.028	> 0.999	0.059	0.059
SD	0.33 (0.11–0.47)	0.39 (0.19–0.58)	0.30 (0.06–0.53)	0.074
Minimum	0.42 (0–1.77)	0.27 (0.01–0.81)	0.28 (0–1.61)	0.644
Maximum	1.70 (1.10–2.65)	1.89 (1.19–2.79)	1.43 (0.87–2.94)	0.125
10th percentile	0.67 (0.04–1.88)	0.59 (0.05–0.87)	0.48 (0.11–1.77)	0.317
25th percentile	0.84 (0.34–1.97)	0.84 (0.48–1.01)	0.66 (0.26–2.16)	0.118
50th percentile	1.04 (0.66–2.08)	1.06 (0.83–1.26)	0.81 (0.49–2.27)	0.048	> 0.999	0.059	0.1689
75th percentile	1.22 (0.98–2.21)	1.39 (0.92–1.63)	0.96 (0.66–2.44)	0.014	> 0.999	0.027	0.041
90th percentile	1.40 (1.02–2.32)	1.56 (1.05–2.25)	1.21 (0.84–2.61)	0.037	> 0.999	0.202	0.038
Kurtosis	2.85 (2.04–8.23)	2.70 (1.71–3.96)	3.10 (2.09–4.91)	0.599
Skewness	−0.13 (−1.92–1.94)	0.28 (−0.63–0.84)	0.11 (−0.94–1.24)	0.084
Entropy	2.69 (0–4.77)	2.59 (1.26–4.68)	3.45 (0–5.6)	0.089

^aCalculated with the Kruskal-Wallis test followed by the Dunn multiple comparisons test.

When comparing D histogram parameters between the three groups, the Kruskal-Wallis test showed significant differences in the mean (P = 0.014), 25th percentile (P = 0.038), 50th percentile (P = 0.011), 75th percentile (P = 0.016), 90th percentile (P = 0.024), and entropy (P = 0.015) of D between them (Table 3). Results of multiple comparison testing showed that the mean (P = 0.017), 25th percentile (P = 0.034), 50th percentile (P = 0.013), 75th percentile, and entropy (P = 0.013) of D in invasive breast cancer were lower than benign lesions. The 90th percentile of D in invasive breast cancer tended to be lower than benign lesions (P = 0.066), but showed a significant difference with that of DCIS (P = 0.040). The 50th percentile (P = 0.059) and 75th percentile (P = 0.053) of D in invasive breast cancer tended to be lower than that of DCIS.

Table 3.

Comparison of DKI (diffusivity and kurtosis) histogram parameters between benign vs. DCIS. vs. invasive breast cancer.

	Benign (n = 23)	DCIS (n = 14)	Invasive (n = 16)	P value	P valuea
					Benign vs. DCIS	Benign vs. Invasive	DCIS vs. Invasive
Diffusivity (D) (10−3 mm2/s)
Mean	1.35 (0.90–2.45)	1.36 (0.98–1.70)	0.98 (0.62–2.91)	0.014	> 0.999	0.017	0.074
SD	0.46 (0.11–0.63)	0.57 (0.24–0.87)	0.42 (0.10–0.66)	0.143
Minimum	0.42 (0–2.05)	0.27 (0.01–0.97)	0.28 (0–1.64)	0.634
Maximum	2.35 (1.24–3.40)	2.37 (1.46–3.35)	1.87 (1.05–4.60)	0.075
10th percentile	0.80 (0.04–2.23)	0.66 (0.05–1.01)	0.53 (0.11–2.23)	0.320
25th percentile	1.07 (0.34–2.31)	1.05 (0.48–1.26)	0.79 (0.26–2.58)	0.038	> 0.999	0.034	0.316
50th percentile	1.35 (0.82–2.45)	1.36 (1.01–1.55)	0.99 (0.44–2.80)	0.011	0.832	0.013	0.059
75th percentile	1.56 (1.15–2.58)	1.71 (1.21–2.34)	1.20 (0.84–3.06)	0.016	> 0.999	0.027	0.053
90th percentile	1.90 (1.22–2.82)	2.00 (1.30–2.83)	1.56 (1.00–3.73)	0.024	> 0.999	0.066	0.040
Kurtosis	3.03 (1.86–7.96)	3.05 (1.64–5.19)	2.92 (2.12–6.13)	0.763	0.769
Skewness	−0.14 (−1.56–1.78)	0.07 (−0.58–1.13)	0.34 (−0.80–2.26)	0.196	0.526
Entropy	1.37 (0–3.69)	1.46 (0.64–2.99)	3.17 (0–5.07)	0.015	> 0.999	0.013	> 0.999
DKI Kurtosis (K)
Mean	0.84 (0.40–1.17)	0.81 (0.61–1.00)	0.83 (0.38–1.39)	0.696
SD	0.17 (0.02–0.47)	0.23 (0.12–0.31)	0.20 (0.05–0.38)	0.133
Minimum	0.52 (0.22–0.80)	0.43 (0.04–0.72)	0.54 (0.05–0.81)	0.253
Maximum	1.16 (0.43–2.52)	1.26 (0.90–1.94)	1.24 (0.53–2.26)	0.834
10th percentile	0.63 (0.34–0.84)	0.50 (0.19–0.75)	0.64 (0.05–0.88)	0.048	0.059	> 0.999	0.136
25th percentile	0.71 (0.36–0.97)	0.63 (0.43–0.88)	0.71 (0.35–1.06)	0.188
50th percentile	0.80 (0.40–1.16)	0.77 (0.61–0.98)	0.83 (0.38–1.39)	0.540
75th percentile	0.94 (0.41–1.30)	0.94 (0.77–1.13)	0.93 (0.40–1.57)	0.745
90th percentile	1.06 (0.42–1.57)	1.03 (0.81–1.45)	1.00 (0.41–1.77)	0.829
Kurtosis	3.44 (1.49–17.87)	3.66 (1.67–11.72)	3.14 (1.57–10.02)	0.962
Skewness	0.34 (−1.88–3.64)	0.40 (−2.70–1.45)	0.17 (−2.49–1.66)	0.560
Entropy	3.81 (1.56–5.82)	4.21 (2.35–5.56)	3.74 (0.64–5.59)	0.521

^aCalculated with the Kruskal-Wallis test followed by the Dunn multiple comparisons test.

When comparing K histogram parameters between the three groups, the Kruskal-Wallis test showed a significant difference in the 10th percentile of K (P = 0.048). When performing multiple comparisons testing, the 10th percentile of K tended to be higher in benign lesions (P = 0.059), but none showed a significant difference.

The AUC for discriminating between benign lesions and invasive cancer ranged from 0.723 to 0.758 (x10⁻³ mm²/s) for the mean, 50th percentile, 75th percentile, and entropy of D, with high values of specificity ranging from 91.3% to 95.7% (Table 4). The AUC for the 75th percentile of ADC was 0.750, with a specificity of 100%. For distinguishing invasive cancer from DCIS lesions, the AUC values were 0.768,0.748 and 0.732 for the 75th of ADC, 90th percentile of ADC and the 90th percentile of D respectively, with lower values of specificity ranging from 71.43% to 85.71%.

Table 4.

ROC curve analysis of ADC, DKI histogram measures which showed a significant difference according to pathology.

Benign vs. Invasive	ADC-75th percentile	D-Mean	D-25th percentile	D-50th percentile	D-75th percentile	D-Entropy
Threshold (10−3mm2/s)	≤ 0.965	≤ 1.065	≤ 0.939	≤ 1.002	≤ 1.234	> 2.581
AUC	0.750	0.755	0.723	0.753	0.750	0.758
Sensitivity (%)	56.2%	62.5%	75.0%	62.5%	56.2%	62.5%
Specificity (%)	100%	91.3%	69.57%	95.7%	95.7%	91.3%
P value	0.005	0.003	0.011	0.004	0.005	0.003

DCIS vs. Invasive	ADC-75th percentile	ADC-90th percentile	D-90th percentile
Threshold (10−3 mm2/s)	≤ 1.297	≤ 1.331	≤ 1.764
AUC	0.768	0.748	0.732
Sensitivity (%)	81.25%	68.75%	75.0%
Specificity (%)	71.43%	85.71%	78.6%
P value	0.004	0.009	0.019

3.3. Correlation between D and K according to pathology

The mean values of D and K showed strong correlation in all lesions (r_s = −0.684, P < 0.001), and very strong correlation in benign (r_s = −0.813, P < 0.001) and invasive breast lesions (r_s = −0.853, P < 0.001) (Fig. 3). There was no significant correlation between mean values of D and K among DCIS lesions (P = 0.589). There was one outlier among the invasive additional breast lesions, which showed a high mean D value of 2.91 (x10⁻³ mm²/s) and a low K value of 0.382. This lesion was confirmed as mucinous carcinoma, which was the same pathology as the index cancer lesion.

Fig. 3.

Scatter plot shows negative correlation between the mean values of D (10⁻³ mm²/s) and K in benign and invasive breast lesions, but no correlation in DCIS lesions.

3.4. Correlation of ADC, D, and K between additional suspicious breast lesions and invasive index breast cancers

Among the 40 additional suspicious lesions that were detected in patients with invasive index breast cancer, the mean values of ADC, D and K showed very strong to strong correlation between invasive additional and index breast cancer lesions (ADC, r_s = 0.815, P < 0.001; D, r_s = 0.823, P < 0.001; K, r_s = 0.744, P = 0.001) (Fig. 4). However, there was no correlation between mean values of ADC, D and K between additional suspicious lesions that were confirmed as benign and index invasive breast cancer lesions (ADC, r_s = 0.256, P = 0.399; D, r_s = 0.275, P = 0.363; K, r_s = 0.143, P = 0.641), or between additional DCIS lesions and index invasive breast cancer lesions (ADC, r_s = 0.077, P = 0.821; D, r_s = 0.292, P = 0.384; K, r_s = −0.009, P = 0.979).

Fig. 4.

Scatter plot shows correlation of mean values of ADC (10⁻³ mm²/s) (a), diffusivity (D) (10⁻³ mm²/s) (b), and kurtosis (K) (c) between the invasive index and invasive additional breast cancer lesions in 16 patients.

4. Discussion

Although unexpected additional suspicious breast lesions are detected in approximately 8%–25% of patients undergoing preoperative MR imaging [26–28], the reported specificity of DCE-MR imaging is relatively variable. To improve its diagnostic performance, several researchers have investigated whether quantitative ADC parameters can aid in predicting additional malignancy in patients with known breast cancer [28,29]. Additional malignant lesions were reported to have lower mean ADC values than benign lesions, with AUC values of mean ADC ranging from 0.76 to 0.81. However, the majority of additional malignant lesions in previous studies were invasive breast cancer, with DCIS accounting for only about 13% [28,29]. In our study, 46.7% of the malignant breast lesions were DCIS and there was no significant difference in mean ADC values between DCIS and benign breast lesions. Therefore, our results may imply that mean values of ADC would be less helpful for evaluating additionally detected lesions on preoperative MRI in patient groups that have a high prevalence of DCIS or when accurate identification of DCIS is critical.

Recently, ADC histogram analysis has also been applied for the characterization of index breast lesions [30–32]. Previous researchers have reported a higher diagnostic performance of minimum ADC than mean ADC in slightly larger studies comprised of 75 to 101 patients, in which DCIS accounted for 4%–14% of index malignant breast lesions [31,32]. In our study, although the minimum ADC did not significantly differ according to pathology, several other ADC histogram parameters showed a significant difference between benign vs. invasive and DCIS vs. invasive cancer, also suggesting that ADC histogram analysis could aid in characterization of additional suspicious lesions at preoperative breast MR imaging.

DKI has recently been investigated as another adjunctive method to improve the performance of breast DCE-MR imaging, mainly by increasing its specificity [19,20]. The reported AUCs of D for the differentiation of benign and malignant breast lesions range between approximately 0.95 to 0.97, respectively [19,20,33,34]. As previous studies had included main breast lesions that were considered suspicious at mammography or US, the average lesion size was larger than that in our study − especially for malignant lesions, which ranged from 24 to 36 mm [19,20,33]. In addition, the majority of malignant lesions in previous studies were invasive breast cancer. Such differences in lesion type and size would be expected, as additionally detected suspicious lesions at preoperative imaging would more likely be smaller and have a higher proportion of DCIS than index cancers [12]. The lower performance of DKI in our study compared to previous reports would be at least partially attributed to differences in the study population [17–20,35]. Furthermore, K histogram parameters failed to show a significant difference at multiple comparisons. Although the application of diffusion gradients in three orthogonal directions to generate trace-weighted images is the standard imaging protocol in body DWI and has been used in previous breast DKI studies [18,19,34], recent research has reported that the use of trace-weighted images can introduce both bias and error in the estimation of DKI-derived indices, especially for K values [36,37]. As it has been shown that water diffusion in breast tissue is not isotropic [38,39], the use of only three diffusion weighting directions would not allow a rotational invariant estimation of diffusion kurtosis. For a rigorous estimation, a protocol including at least 15 diffusion weighting directions and 2 non-null b-values would be required [40]. Furthermore, although we applied a high b value of 3000 based on earlier studies utilizing DKI analysis [18,19,41], this does not necessarily suggest that the b values used were optimal for accurate DKI analysis. These factors may have further affected the performance of DKI analysis in our study. In addition, non-linearity of diffusion gradients and eddy current-induced distortions have been shown to cause inaccuracy in quantitative DWI [42–46], and further improvement in future studies may be achieved by applying gradient non-linearity and eddy current distortion correction.

When performing multiple comparisons between additional benign vs. DCIS vs. invasive breast lesions, we found that the mean ADC values tended to be lower in invasive breast cancer than benign lesions or DCIS lesions. We found that the mean D value in invasive breast cancer was significantly lower than that of benign lesions (P = 0.017) and tended to be lower than that of DCIS (P = 0.074). However, neither differed between benign and DCIS lesions. Histogram analysis showed that multiple D histogram parameters (25th, 50th, 75th and 90th percentile and entropy) and two ADC histogram parameters differed between invasive cancer vs. benign or invasive cancer vs. DCIS. However, none of the histogram parameters showed a significant difference between benign and DCIS lesions. As D histogram parameters that significantly differed according to pathology showed similar diagnostic performance with significant ADC histogram parameters, it may be implied that although DKI analysis can help differentiate between benign and malignant additional lesions detected on preoperative breast MRI imaging, it may have little additional value to conventional ADC analysis.

Similarly, when performing correlation analysis between additional suspicious breast lesions and index invasive breast cancer, we found that although the mean values of ADC, D, and K showed very strong correlation between invasive additional and index breast cancer lesions, there was no correlation between mean values of ADC, D, and K between additional benign/DCIS lesions and invasive index breast cancer. Our results suggest that the histologic type could be the primary factor affecting ADC and DKI parameters. In addition, although the mean values of D and K showed very strong correlation in additional benign and invasive breast lesions, no significant correlation was seen in DCIS. We speculate that possibly due to inherent heterogeneity of DCIS, other factors in addition to non-Gaussian distribution may affect D or K in this subgroup. Our results suggest that when evaluating additional suspicious lesions on preoperative breast MR imaging, DKI may help lesion characterization. Yet, differentiating DCIS from benign lesions would still be difficult and thus, could limit its role in this subgroup. In contrast, we anticipate that DKI could have a higher potential in future non-contrast screening breast MR imaging, by aiding in the detection of small invasive breast cancers that would be more likely clinically significant than small DCIS.

Our study had several limitations. First, this was a single-center study and the number of included patients was small. Second, the retrospective study design and inclusion of lesions larger than 10 mm would have inevitably led to selection bias. Such a size criterion would limit its range of application in clinical practice, as a large portion of additional MR-detected lesions are small and caused us to include only a quarter of initially eligible lesions. However, such a size threshold is a common approach in exploratory studies investigating the potential of diffusion-weighted imaging techniques, and would allow further expansion to smaller lesions in future studies [47,48]. Third, ROIs were manually drawn on a representative slice, and whole-volume lesion analysis was not performed. Although previous studies on DKI have also used this approach, this could have affected results of histogram analysis [18,33]. Fourth, DKI and ADC analysis were based on datasets with different acquisition times. Although this would have allowed comparison with conventional DWI used in clinical practice, it would have caused bias in the comparison between the two diffusion-MRI techniques. Fifth, a specific quality control program for evaluation of MR scanner system-related factors in diffusion indices was not performed prior to the initiation of the study. Such quality assurance procedures have been recently emphasized in order to guarantee accurate and reproducible diffusion measurements, and would be recommended in future diffusion-MRI studies [49–52]. Sixth, imaging processing to correct diffusion-weighted images for eddy current induced distortion was not performed. Finally, our study population had a high proportion of DCIS (46.7% among malignant lesions) compared with previous studies on DKI and DWI imaging. However, this likely represents the characteristics of additional suspicious lesions detected at preoperative breast imaging, and would be informative when considering its application in this subgroup.

5. Conclusion

In conclusion, our study shows that DKI analysis may aid in the differentiation of additional suspicious lesions detected on preoperative breast MR imaging, but may have little additional value to ADC analysis. Both ADC and DKI may have lower potential in differentiating DCIS from benign lesions, and further studies are required to evaluate its role in this subgroup.