3D Texton Based Prostate Cancer Detection Using Multiparametric Magnetic Resonance Imaging

Multiparametric magnetic resonance imaging (mp-MRI) has shown its potential in prostate cancer detection. In this study, we investigate the application of 3D texton based prostate cancer detection using T2-weighted (T2W) MRI, dynamic contrast-enhanced (DCE) MRI and apparent diffusion coefficient (ADC) maps. For the T2W and ADC modalities, the traditional texton based approach is adopted, i.e., for each voxel, a texton histogram is extracted as the feature to perform the classification. For the DCE data, we present a new method, where the textons are extracted from each series and for each voxel, the corresponding textons across all series are used as features. A random forest classifier is applied for classifying all voxels into benign or malignant. The evaluation is conducted by performing a receiver operating characteristics (ROC) analysis and computing the area under the curve (AUC). The experiments on the Initiative for Collaborative Computer Vision Benchmarking (I2CVB) database demonstrate that the texton based approach using mp-MRI data obtains excellent performance in prostate cancer detection and produces 88.3% accuracy, whereas the accuracy produced by an intensity based approach is 79.8%.