Diffusion Kurtosis Breast Imaging model – Which should be the highest b-value?

Filipa Borlinhas^{1}, Luísa Nogueira^{2}, Sofia Brandão^{3}, Rita G. Nunes^{1}, Raquel Conceição^{1,4}, Joana Loureiro^{3}, Isabel Ramos^{3}, and Hugo A. Ferreira^{1}

Diffusion Kurtosis Imaging (DKI) can
be used to model Diffusion Weighted MRI data so as to estimate how much the distribution
probability of water displacement deviates from a Gaussian function; this is
due to the presence of tissue barriers. This model includes the Mean Diffusivity (MD) parameter, which corresponds to the Apparent
Diffusion Coefficient when the water displacement distribution probability is
Gaussian, and the Mean Kurtosis (MK), which reflects the deviation from Gaussianity^{1,2}. Recently, it has been stated that
these parameters can be used to differentiate benign and malignant breast lesions^{3,4}.
In DKI studies, the use of high b-values is recommended^{1} but the optimal
highest b-value in breast DKI remains uncertain.

The purpose of this study was to
assess which high b-value, 3000 or 2000s/mm^{2}, for calculating the DKI parameters so as to
differentiate benign and malignant breast lesions.

When higher b values are used MD and MK mean values are lower, considering both lesion types (Table 1). Considering any of the b-value combinations under study, both DKI parameters can differentiate between benign and malignant breast lesions (p=0.000) (Table 1). The Wilcoxon Signed-Rank test indicates (Table 2) that there are no significant differences between b-value groups in benign tumors, but significant differences are present considering the malignant ones. This means that the choice of used b-values range is very important.

Considering the ROC curve analysis (Figure 1), the b-value range with the best performances
is b=50 to 3000s/mm^{2}, for both MD and MK parameter.
Moreover, for that b-value range, higher sensitivity was obtained (0.77 and 0.79 for MD and
MK, respectively) (Table 3), indicating a stronger potential to identify the presence of malignant
lesions. Likewise, the
highest specificity was obtained for the same b-value range with the MD
parameter (0.89).

Using the b=50 to 3000s/mm^{2},
the highest accuracies were obtained for both parameters (0.81 for MD and 0.80
for MK) (Table 3), the higher values for Negative Predictive Value (NPV) were
obtained for both DKI parameters (0.65), and the higher Positive Predictive
Value (PPV) was obtained for MD parameter (0.94). These results indicate that these b-values are more suitable
when applying the DKI
model.

Furthermore, as shown in Table 3, the
lower False Negative Rate (FNR) was obtained for MK (0.21), and the lower False
Positive Rate (FPR) was obtained for MD parameter (0.11), in both cases using
the 50 to 3000s/mm^{2} b-value range. Consistently, the risk of having a False
Negative or a False Positive result is higher for the b-value group ranging
from 50 to 2000s/mm^{2}.
The RMSE associated to model fitting
resulted in similar values for the two methods considered (2.04 for b=3000s/mm^{2}, and 1.96 for b=2000 s/mm^{2}).

These experiments may indicate the
need for using b-values as high as 3000s/mm^{2} in breast DKI.
However, since using 2000s/mm^{2} as the highest b-value, also enables
distinguishing breast lesions, and as the high b-value images are noisier,
care is needed when interpreting these results.

The two b-value ranges under study enabled
the differentiation of benign and malignant breast lesions. The majority of the
statistical tests performed suggest that the b-value range from 50 to 3000s/mm^{2} is the best one for breast DKI from
those tested. Further tests are needed to evaluate if the same result would be observed
when considering the differentiation between malignant lesions subtypes.

Future studies applying DKI to
breast imaging should take this result into consideration and include a high b-value
of 3000s/mm^{2}.

1. Jensen JH, Helpern JA, et al. Diffusional kurtosis imaging: the quantification of non-gaussian water diffusion by means of magnetic resonance imaging. Magn Reson Med. 2005 Jun;53(6):1432–40.

2. Jensen JH, Helpern J. MRI quantification of non-Gaussian water diffusion by kurtosis analysis. NMR Biomed. 2010;23(7):698–710.

3. Nogueira L, Brandão S, et al. Application of the diffusion kurtosis model for the study of breast lesions. Eur Radiol. 2014 Jun;24(6):1197–203.

4. Wu D, Li G, et al. Characterization of breast tumors using diffusion kurtosis imaging (DKI). PLoS One. 2014;9(11):e113240.

Figure 1 - On the left, the ROC curves for MD
are represented, and on the right the ROC curves for MK, considering the 2
b-value ranges. The pink and the yellow curves are the ones with the
best performances, which correspond to the 50 to 3000s/mm^{2} b-value range.

Table 1 – Mean±SD, maximum and minimum values
for benign and malignant breast tumors, obtained for DKI parameters estimated using the
2 combinations of b-values (b=50 to 2000s/mm^{2}, b=50 to 3000s/mm^{2}). b-values - s/mm^{2}, SD - Standard Deviation, MD - Mean Diffusivity (×10^{-3}mm^{2}/s),
MK – Mean Kurtosis (dimensionless measure), *-significant
differences.

Table 2 – Results for Wilcoxon Sign Test and
the respective p-values, for the different b-value groups considered in this
study. MD - Mean
Diffusivity (×10^{-3}mm^{2}/s), MK – Mean Kurtosis, * -
statistical significant differences.

Table 3 – Statistical parameters extracted for
each range of b-value in study. b-values - s/mm^{2}, MD -
Mean Diffusivity(×10^{-3}mm^{2}/s), MK – Mean Kurtosis,
Area Under the Curve (AUC), Confidence Intervals (CI), Positive Predictive
Value (PPV), Negative Predictive Value (NPV), False Negative (FN) rate ,
False Positive (FP) rate, * - significant differences.

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

3017