A Comprehensive Evaluation of Quantitative Diffusion Parameters for Differentiating Histopathological Features and Subtypes of Breast Cancers: Diffusion Kurtosis Imaging (DKI), Intravoxel Incoherent Motion (IVIM) and Histogram Analysis of ADC

Abstract

Background The objective of this study is to quantitatively compare the diagnostic value of conventional diffusion-weighted imaging (DWI), intravoxel incoherent motion (IVIM), and diffusion kurtosis imaging (DKI) in differentiating the histopathological features and subtypes of breast cancer. Materials and Methods There were 98 patients with breast cancer studied by multiple b value DWIs and DKIs grouped according to their molecular prognostic factors. Entropy and histogram derived parameters of volumetric ADC values, true diffusivity (Dt), pseudo-diffusion coefficient (Dp), perfusion fraction (f), mean kurtosis (MK), and mean diffusivity (MD) maps were calculated using voxel based analysis for the whole lesion volume. The diagnostic efficacy of various diffusion parameters for predicting both molecular prognostic factors (Hormone-Receptor (HR, ER or PR positive), HER2 and ki67) and breast cancer subtypes were compared. Diagnostic performance was evaluated using the univariate and multivariate logistic regressions, ROC analysis, multivariate backward logistic regression, analysis of covariance (ANCOVA) and partial eta squared (ηp2) estimation. Results HR- positive tumors had significantly lower median ADC values (P= < 0.001, Bonferroni adjusted significance < 0.002) than HR- negative tumors. HER-2 positive tumors had significantly higher mean ADC values and last ADC quartile (P< 0.001, univariate regression: OR=99.3, 14.2, AUC=0.79, 0.73, P<0.001) than HER-2 negative tumors. High ki67 tumors had significantly lower last ADC quartile (P< 0.001) than tumors with low ki67 index. Luminal B subtype had significantly lower mean ADC, median ADC (OR=0.011, AUC=0.78, P<0.001) and last ADC Quartile (P< 0.001, Bonferroni adjusted significance < 0.001), HER-2 subtype had significantly higher mean ADC, median ADC and last ADC Quartile (P< 0.001, (OR=129.2, 32.1, 78.7, univariate regression, P<0.001, AUC=0.94, 82, 89, P<0.001) and triple negative subtype showed significantly lower MD (P< 0.001, univariate regression: OR=0.02, AUC=0.73, P=0.002) than other tumor subtypes. ANCOVA analyses found a significant association between mean ADC and luminal HER2 (ηp2=0.86, P< 0.001) after adjusting for molecular prognostic factors. Conclusion The use of diffusion imaging with multiple b values will be beneficial for the classification of breast cancers.