Improved detection and characterization of obscured central gland tumors of the prostate: texture analysis of non contrast and contrast enhanced MR images for differentiation of benign prostate hyperplasia (BPH) nodules and cancer

Abstract

OBJECTIVE: The purpose of this study to assess the value of texture analysis (TA) for prostate cancer (PCa) detection on T2 weighted images (T2WI) and dynamic contrast-enhanced images (DCE) by differentiating between the PCa and Benign Prostate Hyperplasia (BPH). MATERIALS & METHODS: This study used 10 retrospective MRI data sets that were acquired from men with confirmed PCa. The prostate region of interest (ROI) was delineated by an expert on MRI data sets using automated prostate capsule segmentation scheme. The statistical significance test was used for feature selection scheme for optimal differentiation of PCa from BPH on MR images. In pre-processing, for T2-WI, Bias correction and all images intensities are standardized to a representative template. For DCE images, Bias correction and all images are registered to time point 1 for that patient. Following pre-processing texture, features from ROI were extracted and analyzed. Texture features that were extracted are: Intensity mean and standard deviation, Sobel (Edge detection), Haralick features, and Gabor features. RESULTS: In T2-WI, statistically significant differences were observed in Haralick features. In DCE images, statistically significant differences were observed in mean intensity, Sobel, Gabor, and Haralick features. CONCLUSION: BPH is better differentiated in DCE images compared to T2-WI. The statically significant features may be combined to build a BPH vs. cancer detection system in future.