Learning Approach to Analyze Tumour Heterogeneity in DCE-MRI Data During Anti-cancer Treatment

Abstract

Abstract. The paper proposes a learning approach to support medical researchers in the context of in-vivo cancer imaging, and specifically in the analysis of Dynamic Contrast-Enhanced MRI (DCE-MRI) data. Tumour heterogeneity is characterized by identifying regions with different vascular perfusion. The overall aim is to measure volume differences of such regions for two experimental groups: the treated group, to which an anticancer therapy is administered, and a control group. The proposed approach is based on a three-steps procedure: (i) robust features extraction from raw time-intensity curves, (ii) sample-regions identification manually traced by medical researchers on a small portion of input data, and (iii) overall segmentation by training a Support Vector Machine (SVM) to classify the MRI voxels according to the previously identified cancer areas. In this way a non-invasive method for the analysis of the treatment efficacy is obtained as shown by the promising results reported in our experiments.