Quantitative analysis of diffusion-weighted magnetic resonance imaging in malignant breast lesions using different b value combinations
- First Online:
- Cite this article as:
- Nilsen, L.B., Fangberget, A., Geier, O. et al. Eur Radiol (2013) 23: 1027. doi:10.1007/s00330-012-2687-8
- 1.2k Downloads
To explore how apparent diffusion coefficients (ADCs) in malignant breast lesions are affected by selection of b values in the monoexponential model and to compare ADCs with diffusion coefficients (Ds) obtained from the biexponential model.
Twenty-four women (mean age 51.3 years) with locally advanced breast cancer were included in this study. Pre-treatment diffusion-weighted magnetic resonance imaging was performed using a 1.5-T system with b values of 0, 50, 100, 250 and 800 s/mm2. Thirteen different b value combinations were used to derive individual monoexponential ADC maps. All b values were used in the biexponential model.
Median ADC (including all b values) and D were 1.04 × 10-3 mm2/s (range 0.82–1.61 × 10-3 mm2/s) and 0.84 × 10-3 mm2/s (range 0.17–1.56 × 10-3 mm2/s), respectively. There was a strong positive correlation between ADCs and Ds. For clinically relevant b value combinations, maximum deviation between ADCs including and excluding low b values (<100 s/mm2) was 11.8 %.
Selection of b values strongly affects ADCs of malignant breast lesions. However, by excluding low b values, ADCs approach biexponential Ds, demonstrating that microperfusion influences the diffusion signal. Thus, care should be taken when ADC calculation includes low b values.
• Diffusion-weighted sequences are increasingly used in breast magnetic resonance imaging
• Diffusion-weighting (b) values strongly influence apparent diffusion coefficients of malignant lesions
• Exclusion of low b values reduces the apparent diffusion coefficient
• Flow-insensitive monoexponential apparent diffusion coefficients approach biexponential diffusion coefficients
KeywordsApparent diffusion coefficient (ADC)Monoexponential modelBiexponential modelBreastDiffusion-weighted magnetic resonance imaging
Several potential applications for the ADC in breast cancer have been suggested and studied; including detection, characterisation, differentiation of tumours as well as evaluation of neoadjuvant treatment response [1, 2, 4–8]. ADC has been found to correlate with cell density and studies have reported increasing ADC during neoadjuvant chemotherapy, reflecting reduced cellularity due to cell damage and/or death [6–8]. Currently, the most unambiguous results have been found when ADC is used to differentiate between malignant and benign breast lesions . Including ADC in the characterisation of tumours has been shown to increase the diagnostic accuracy compared with dynamic contrast-enhanced MRI alone [9–11]. However, ADC of malignant and benign breast lesions has been shown to overlap, both within and between different studies, and there is no established cut-off value. Reported ADCs of malignant and benign lesions vary from 0.68 to 1.61 × 10-3 mm2/s [12, 13] and from 1.35 to 1.77 × 10-3 mm2/s [14, 15], respectively, whereas previously reported ADCs of normal breast tissue range from 1.51 to 2.09 × 10-3 mm2/s [16, 17]. Different b value combinations and calculation schemes as well as imaging-dependent factors  are anticipated to contribute to the large discrepancy in published ADCs.
Traditionally, breast ADCs have been calculated using the monoexponential model. But, there is no consensus on an optimal b value combination and generally b = 0 s/mm2 is included. The purpose of this study was to examine how the choice of b values included in the monoexponential model affects calculated ADCs and also to compare these with Ds derived from the biexponential model.
Materials and methods
Twenty-four women diagnosed with locally advanced breast cancer were included in this prospective study. The mean age at time of inclusion was 51.3 years (range 37–72 years). Eighteen lesions were categorised as invasive ductal carcinomas (grade 2, n = 7; grade 3, n = 8; not determined, n = 3) and 6 lesions as invasive lobular carcinomas (grade 2, n = 4 and grade 3: n = 2). The mean initial clinical tumour size (longest tumour diameter) of the breast lesions was 6.7 cm (range 5–11 cm). Written consent was obtained from all women and the study was approved by the regional ethics committee and the protocol review committee of our institution.
The MR examination was performed as part of the radiological pre-treatment assessment of the extent of the disease. Patients underwent imaging in the prone position using a dedicated phased-array bilateral breast coil (CP Breast array coil, Siemens) on a 1.5-T MR system (ESPREE, Siemens, Erlangen, Germany). To reduce motion artefacts during imaging, cotton pads were put inside the breast coils and care was taken to avoid compression of the breasts.
Diffusion-weighted MRI was carried out using a single-shot spin-echo echo-planar imaging sequence with fat-saturated short T1 inversion recovery and b values of 0, 50, 100, 250 and 800 s/mm2 in three orthogonal directions, and the total imaging time was 6:52 min. The following DW MRI imaging parameters were used: repetition time (TR) = 10,300 ms; echo time (TE) = 126 ms; inversion time (TI) = 190 ms; slice thickness = 4 mm; slice gap = 2 mm; number of slices = 26; number of excitations (NEX) = 3; field of view = 360 mm × 195 mm; image matrix = 192 × 104; echo spacing = 0.94 ms; bandwidth = 1,240 Hz/pixel; phase encoding from anterior to posterior. The MR examination also included sagittal turbo spin-echo T1-weighted MRI, axial turbo spin-echo T2-weighted MRI and T1-weighted 3D dynamic contrast-enhanced (DCE) MRI with gadopentetate dimeglumine (Magnevist, Schering, Berlin, Germany) as contrast agent (0.1 mmol/kg body weight). The DCE MRI included acquisition of one pre-enhanced and four enhanced sequences with a time resolution of 85 s. DW MRI was acquired before contrast agent injection.
A region of interest (ROI) was drawn manually within a solid part of each breast lesion on an ADC map calculated using all b values in the commercially available nICE software package (Nordic NeuroLab, Bergen, Norway). The drawing of the ROI was guided by the native b800 image and the DCE MRI subtraction image (second postcontrast DCE MRI image – precontrast DCE MRI image) from the same anatomical location as the ADC map. The placement of the ROI was verified by an experienced breast radiologist. Further calculations of ADCs, Ds, D*s and fps were performed for all ROIs on a voxel-by-voxel basis using in-house written IDL routines (Interactive Data Language [IDL] version 6.3; Research System, Boulder, CO, USA). In each voxel of the ROI, the DW MRI SI as a function of the b value was fitted to the mono- and biexponential models, using Levenberg–Marquardt least squares minimisation implemented in the commercially available IDL procedure MPFIT (Markwardt CB, presented at the 2008 conference Astronomical Data Analysis Software and Systems XVIII). The goodness of fit between measured SI and fitted curves was evaluated by the Pearson’s correlation coefficient squared (r2) and the summed squared residuals of the fit (χ2).Voxels where the curve fitting resulted in low r2 or high χ2 were excluded from further analysis.
The different b value combinations (BXsa) explored in the study
b values [s/mm2]
0, 50, 100, 250, 800
50, 100, 250, 800
100, 250, 800
0, 50, 100, 250, 800
All statistical analysis was performed using SPSS version 18.0 (SPSS, Chicago, IL, USA). Ds and ADCs calculated using the different b value combinations were compared using the non-parametric Wilcoxon signed-rank test. The non-parametric Spearman’s ρ correlation was used to evaluate correlations between Ds and different ADCs. The significance level was set at 5 %.
Comparison of monoexponentially and biexponentially calculated diffusion coefficients: results of calculation with B0–B11 and B0biexp
Mean ± SD [×10-3 mm2/s]
Median [×10-3 mm2/s]
Range [×10-3 mm2/s]
95 % CI [×10-3 mm2/s]
1.12 ± 0.24
1.16 ± 0.23
1.10 ± 0.24
1.09 ± 0.24
1.07 ± 0.25
1.08 ± 0.24
1.03 ± 0.25
2.32 ± 0.65
1.89 ± 0.41
1.51 ± 0.30
1.55 ± 0.41
1.32 ± 0.26
0.90 ± 0.30
Comparison of monoexponentially and biexponentially calculated diffusion coefficients: deviation [%] and Spearman’s ρ correlation of median values for B0-B5 and B0biexp
Correlation: DB0biexp and ADCB0-B5
The median biexponential D* was 11.9 (range 4.3–72.1) and median fp was 16 % (range 5–55 %). There was no correlation between D* and ADCs calculated using the three lowest b values ADCB7–B8.
This study shows that there is significant variability in monoexponentially calculated breast tumour ADCs when different b value combinations are used. However, for clinically used b value combinations (B0–B5), the maximum deviation is only 11.8 % (Table 3). The variability in calculated ADCs is reduced when low b values are omitted. Although ADCs are overestimated compared with biexponentially calculated Ds, there is a strong correlation between these two parameters.
Breast tumour ADCs for clinically used b value combinations in our study (mean 1.07–1.16 × 10-3 mm2/s, median 0.97–1.09 × 10-3 mm2/s) are within the range of previously reported values (0.68–1.61 × 10-3 mm2/s) [12, 13]. Consequently, for single-institution monitoring of individual treatment response using ADC, the b value combination may be of limited value, especially if low b values are excluded. But, clinical use of ADC cut-off values for tissue differentiation requires standardisation of acquisition protocol.
It has been suggested that the choice of b value combination have limited impact on the differentiation between malignant and benign breast lesions as studies have shown equal diagnostic performance of monoexponential flow-insensitive ADCs and ADCs calculated including low b values [12, 19, 20]. However, Ds from the biexponential model have been shown to provide better differentiation than ADCs . Furthermore, including b values >1,000 s/mm2 in the ADC calculation might improve the differential diagnostic accuracy  as this reduces the contamination from microperfusion and ADC approaches the biexponentially derived D. But, if the b value is too high for the available signal-to-noise ratio the ADC will be underestimated. In our study, flow-insensitive ADCs of malignant breast lesions were correlated with biexponentially calculated Ds (P < 0.01), but overestimated by about 23 % (Table 3), which is in accordance with previously published results . This may reflect the high vascularity and high cellularity of these lesions; further radiological–histopathological correlation analyses are needed. The microperfusion effect in normal breast tissue has been shown to be limited, yielding similar ADCs and Ds [21, 22], and thereby supporting the validity of the biexponential model.
A limitation to our study is the small and homogeneous study population, consisting of patients with large malignant tumours (mean 6.7 cm). Although our study included two different tumours types, i.e. invasive lobular (25 %) and ductal (75 %) carcinomas, there were no apparent differences between these types with respect to growth pattern, ADCs or contrast-enhancement characteristics on DCE MRI.
The large inter-patient variations in monoexponential ADCs calculated from only low b values found in this study may be attributed to microperfusion heterogeneity, a feature being of great importance in the exploration of anti-angiogenic treatment . It has been hypothesised that D* and fp may reflect microvessel perfusion. Weak correlations have been reported between fp and initial enhancement on DCE MRI . But in accordance with our results, the SDs of fp and D* were large , and the clinical use of these parameters is questionable. The biological relevance of the biexponential model is not established as there is no clear evidence that the modelled compartments correspond to intravascular and extravascular flow and diffusion. Thus, other mathematical models may be more suitable for describing the DW MRI SI attenuation .
In our study the biexponential model provided poorer fitting than did the monoexponential model. This is probably a result of the limited number of low b values included. Since at least four DW MRI series with different b values are required in the biexponential model, the accuracy of the derived parameters is anticipated to increase with increasing number of b values. Despite the limited number of b values used in our study, calculated Ds, D*s and fps are comparable to those obtained in a study using several more b values . The selected number of b values causes a trade-off between increased precision in the estimated diffusion parameters and acquisition time.
In this study, median and mean diffusion values from single ROIs, placed within a solid part of the lesions, were used. This was done to minimise the influence of partial volume effects and limited spatial resolution. The differences between median and mean ADCs were found to be to the same order as the differences between ADCs calculated using the different b value combinations (B0–B5). Histogram analysis including the entire lesion would perhaps depict lesion heterogeneity better. However, mean values obtained from single ROIs have been used in previously published clinical DW MRI studies of breast cancer .
In conclusion, this study demonstrates that using different b value combinations strongly influences the calculated ADCs. Removing low b values reduces the perfusion contamination of the diffusion signal and ADCs approach the biexponentially calculated Ds. Thus, care should be taken when ADC calculation includes low b values. Future studies involving a larger number of patients should be undertaken in order to further validate these findings.
This work has been supported by grants received from the South-Eastern Regional Health Authority (grant 20100029), The Norwegian Cancer Society (grant 80114001), and The Research Council of Norway (183622/S10). The authors wish to acknowledge Siemens Medical Systems for the provision of the DW MRI sequence. The flow-insensitive ADCs of the present study have been used in two other studies [4, 5].
This article is distributed under the terms of the Creative Commons Attribution License which permits any use, distribution, and reproduction in any medium, provided the original author(s) and the source are credited.