Examination and evaluation of MR radiomics features for characterization of dominant intraprostatic lesions.

Document Type

Conference Proceeding

Publication Date


Publication Title

Cancer Res


Purpose: This pilot study investigates a set of radiomics features extracted from fast relaxation fast spin echo (FRFSE) T2 pulse sequences for normal tissue and Dominant Intraprostatic Lesions (DILs) in twenty prostate cancer patients. Material and Methods: Twenty patients with prostate cancer were studied. All patients had axial FSRFSE T2 scans using a 3 Tesla scanner. A radiologist interpreted MR examinations, and contoured the suspicious DIL and the contralateral section of the prostate gland (normal) on the T2 weighted MR images. Patients underwent a 14-core transrectal Ultrasound Guided Biopsy and localization of positive cores, Gleason score and clinical tumor stage were recorded. 167 radiomics features were extracted from normal and DIL zones. These features were categorized into 8 different sets as following: Intensity Histogram Based (IHB), Gray Level Run Length (GLRL), Law's Textural Information (LAWS), Discrete Orthonormal Stockwell Transform (DOST), Local Binary Pattern (LBP), Two Dimensional Wavelet Transform (2DWT), Two Dimensional Gabor Filter (2DGF), and Gray Level Co-Occurrence Matrix (GLCM) with 8, 7, 18, 18, 6, 48, 40, and 22 features in each category respectively. A Welch's test and the Fisher method were used to test for significant differences among the 167 radiomics features and their subcategories. For all patients, correlation coefficients between the extracted features in the normal and DIL zones were also calculated. Results: According to the Fisher combined p-values, among the eight categories of radiomics features, only 5 feature categories showed a significant difference (IHB, GLRL, DOST, LBPF and GLCM with pFisher= 2.0×10-6, 0.02, 12 ×10-4, 3.7 ×10-3, and 1.5 ×10-6 respectively). Among all 167 features, only 7 showed a significant difference (D=100x[DIL/NP-1]) and small correlation between normal and DIL zones: IHB-Skewness (r=0.19, p=0.03, and D=50.3%), GLCM-Contrast (r=0.12, p=0.03, and D=-67.5%), GLCM-Dissimilarity (r=0.12, p=0.01, and D=-67.5%), GLCM-Entropy (r=0.07, p=0.01, and D=-67.1%), GLCM-Difference-Variance (r=0.12, p=0.01, and D=-67.1%), GLCM-Difference-Entropy (r=0.10, p=0.01, and D=-60.4%), and GLCM-Information-Measure-of-Correlation (r=0.25, p=0.01, and D=-65.1%). Conclusion and Discussion: This pilot study demonstrates the feasibility of using radiomics features from MR images to characterize DILs in prostate cancer patients. Among 167 radiomics features extracted from axial MR T2 FRFSE, 7 features were shown to be potentially significant for distinguishing normal tissue from DILs. This research supports an integrated decision making system, combining clinical factors and radiomics features extracted from MR images, for increasing the DIL detection performance in prostate cancer studies.





This document is currently not available here.