Prostate Cancer Detection Using Image-Based Features in Dynamic Contrast Enhanced MRI

Dynamic Contrast Enhanced Magnetic Resonance Imaging (DCE MRI) provides valuable information in prostate cancer detection. Existing computer-aided detection methods focus on estimating the DCE curves as pharmacokinetic models and directly calculating the perfusion-related measurements from the DCE signals. Substantial image content contained in DCE MRI series, which captures the spatio-temporal pattern receives less attention. This work aims to investigate the performance of the image-based features extracted from DCE MRI on prostate cancer detection. Various image-based features are extracted from DCE MRI series. Their performance on prostate cancer detection is compared with features extracted from the pharmacokinetic models and the perfusion-related measurements. Features are concatenated and feature selection is applied to reduce the feature dimensionality and improve cancer detection performance. Evaluation is based on a publicly available dataset. Using image-based features outperforms using either the features extracted from the pharmacokinetic models or the perfusion-related measurements. By applying feature selection to the aggregation of all features, the performance of prostate cancer detection achieves 0.821, for the area under the receiver operating characteristics curve. This study demonstrates that compared with the commonly used pharmacokinetic models and the perfusion-related features, image-based features provide an additional contribution to prostate cancer detection and can potentially be used as an alternative approach to model DCE MRI.