Semi-quantitative analysis of prostate perfusion MRI by clustering of pre and post contrast enhancement phases

Computer-based analysis is highly effective in information retrieval from Dynamic Contrast Enhance Magnetic Resonance Images (DCE-MRI) for prostate cancer recognition. Quantification and modelling of perfusion curves to extract higher order informative features for use in classification algorithms, is a major step in DCE-MRI analysis where inter-scan independent semi-quantitative models are highly desirable. This paper presents a semi-quantitative analysis model for prostate DCE-MRI, that is independent of inter-scan intensity variations, where the significance of the derived features is maximised by preserving the original shape of the perfusion curve to a maximum degree. The proposed model is evaluated on three different classifiers and 82 annotated prostate peripheral zone lesions in 3T DCE-MRI datasets of 40 patients. Random Forest Classifier yields a promising accuracy of 97.6% and a sensitivity of 92.3%.