Super Voxel Clustering for DCE-MRI Pharmacokinetic Analysis and the Effect on Prediction of Breast Cancer Therapy Response

Motivation: Using super voxels (SVs) to average DCE-MRI time courses from localized regions before pharmacokinetic (PK) modeling may reduce the effects from noise and motion while still capturing tumor heterogeneity. Goal(s): Evaluate how the use of SVs affects the predictive performance of Ktrans for breast cancer (BC) response to neoadjuvant chemotherapy (NAC). Approach: 26 BC patients treated with NAC underwent longitudinal DCE-MRI, which were fitted with the Tofts model and Shutter-Speed model (SSM) using a voxel-wise and SV-based approach. Results: Combining the SV approach with SSM can substantially improve Ktrans predictive performance. Impact: Use of SVs to stabilize advanced DCE-MRI PK models may potentially improve accuracy of estimated Ktrans and its predictive performance for BC response to NAC.