In determining the effectiveness of chemoembolization in HCC, functional MRI has been shown to differentiate responders and non-responders earlier than anatomical measurements such as RECIST or EASL criteria. In previous studies, multiparametric response criteria based on thresholds of changes in ADC and venous enhancement (VE) intensities were proposed. We present improved stratification based on machine learning and image-based features. On a set of 57 chemoembolization patients, the proposed approach achieved a mean classification accuracy of 84% versus 66% for the previous threshold-based approach. These results further demonstrate the incremental value of functional MRI over traditional anatomical measures.