Advanced diffusion-weighted MRI of breast cancer: response to neoadjuvant chemotherapy and correlation with dynamic contrast-enhanced MRI

Background:
Previous studies showed promising applications of intravoxel incoherent motion (IVIM) and stretched-exponential (SEM) models of diffusion-weighted imaging (DWI) in breast imaging; however, their ability to predict early breast cancer response to neoadjuvant chemotherapy (NACT) was minimally investigated.
Aims:
To evaluate accuracy, bias, precision, and in-vivo repeatability of IVIM parameters estimated using different curve-fitting methods and determine the optimum for analysing the acquired clinical breast DWI data. To investigate the value of conventional monoexponential versus advanced (IVIM and SEM) DWI models parameters estimated from whole-tumour, tumour diffusion cold-spot, and perfusion hot-spot regions to assess early breast cancer response to NACT. To explore relationships between IVIM and dynamic contrast-enhanced (DCE)-MRI perfusion-related parameters, and between DWI diffusion coefficients and DCE-MRI cellularity-related measures in the same three tumour regions.
Materials:
MRI dataset of primary breast cancer patients acquired at pretreatment and after one and three NACT cycles. Simulated data represent IVIM parameter ranges observed in these patients.
Results:
Constrained oversegmented-fitting was the optimum IVIM curve-fitting method, producing parameter estimates with the smallest errors, highest precision, and best repeatability. Tumour volume was significantly larger in non-responders across all time-points and demonstrated reasonable predictive performance (AUC=0.84-0.88; p<0.05). The monoexponential model was unable to predict response (p>0.05), while IVIM and SEM models differentiated response groups at pretreatment tumour hot-spot regions and after one NACT cycle in three tumour regions, displaying reasonable predictive performance (AUC=0.71-0.79 at pretreatment, 0.71-0.83 after one cycle; p<0.05). IVIM and DCE-MRI perfusion-related parameters were uncorrelated (p>0.5), but statistically significant, moderate between-subject (r=0.405-0.461; p<0.05) and within-subject (rrm=0.514-0.619; p<0.05) correlations between diffusion coefficients and DCE-MRI cellularity-related measures were observed in the whole-tumour regions.
Conclusion:
IVIM and SEM models demonstrated better predictive capabilities for response than the monoexponential model. While IVIM and DCE-MRI perfusion-related parameters were uncorrelated, diffusion coefficients and DCE-MRI cellularity-related measures correlated.