Examination of zone-based radiomic features for characterization of dominant intraprostatic lesions using MR multi-modal information.

Document Type

Conference Proceeding

Publication Date


Publication Title

Cancer Res


Purpose: We investigated radiomic features extracted from dominant intraprostatic lesions (DILs) of the peripheral zone (PZ) and central gland (CG) from MR multi-modal images of 20 patients with prostate cancer (PCa). Remaining prostate gland (RPG) were included in the analysis. Material and Methods: 20 biopsy-proven PCa patients with no prior radiation treatment were studied. Axial T2 weighted images (T2WI) and diffusion weighted images (DWI) were acquired of the pelvis using a 3T MR scanner. ADC maps were constructed from DWIs. Region of interests delineating DILs and RPGs were contoured on each MR modality. 168 radiomic features were extracted from DIL and RPG volumes (15 pairs from PZ and 5 pairs from CG). Radiomic features were categorized into 8 different sets: Intensity Based Histogram (IBH, 9 features), Gray Level Run Length (GLRL, 7 features), Law's Textural Information (LAWS, 18 features), Discrete Orthonormal Stockwell Transform (DOST, 18 features), Local Binary Pattern (LBP, 6 features), 2D Wavelet Transform (2DWT, 48 features), 2D Gabor Filter (2DGF, 40 features), and Gray Level Co- Occurrence Matrix (GLCM, 22 features). ANOVA (with Bonferroni adjustment), overall mean percent difference (OMPD), and the Fisher combined probability were used to test the following 7 hypotheses: (1) DILs of PZ and CG from T2WI (2) DILs of PZ and CG from ADC (3) DILs and RPGs of PZ from T2WI (4) DILs and RPGs of PZ from ADC (5) DILs and RPGs of CG from T2WI (6) DILs and RPGs of CG from ADC (7) DILs of PZ for T2WI and ADC. Results: Results imply that among 168 radiomics features, only 5 (DOST, and 2DGF, OMPD=107.8%), 2 (2DWT, and 2DGF, OMPD=141.7%), 13 (IBH, 2DWT, and GLCM, OMPD=%226.6), 17 (IBH, 2DWT, and GLCM, OMPD=179.7%), 18 (IBH, 2DWT, and GLCM, OMPD=321.9%), 18 (IBH, 2DWT, and GLCM, OMPD=726.1%), and 74 (IBH, GLRL, LAWS, DOST, LBP, 2DWT, 2DGF, GLCM, OMPD=1564%) features are discriminant (p <0.050 with Confidence Level of 95%) for hypotheses no. 1 through 7 respectively. Conclusion and Discussion: Results for the discriminant features identified from hypotheses no. 1 and 2 can be used to construct a predictive model with a higher performance that benefits from the zone-based information (PZ and CG) as a-priori knowledge. As most of the discriminant features in tests no. 1 to 7 are primarily entropy-based, this suggests that potential feature-based biomarkers of DILs in PCa patients are more associated with spatial-locality, and frequency-based characteristics of MR images. The high value of the OMPD for test no. 7, strongly supports the use of the two MR modalities for increasing the information gain in perfecting predictive models for detection of DILs in PCa studies. The results of this pilot study, albeit subject to confirmation in a larger patient population, suggest a potential role for the use of zone-based radiomics information in models developed for detection of DILs in PCa patients.





This document is currently not available here.