Program Information
Assessment and Localization of Aggressiveness in Prostate Cancer Using Dynamic Contrast Enhanced-MRI
N A Parra*, F Chinea , S Punnen , M Abramowitz , A Pollack , R Stoyanova , University of Miami, Miami, FL
Presentations
MO-DE-207B-2 (Monday, August 1, 2016) 1:45 PM - 3:45 PM Room: 207B
Purpose: To automatically assess prostate cancer aggressiveness and to map areas of high risk based on quantitative features from Dynamic Contrast Enhanced-MRI (DCE-MRI).
Methods: Fifty-seven patients with DCE-MRI acquired on 3T Discovery MR750 GE were retrospectively analyzed. Prostate, peripheral zone (PZ) and gluteus maximus (DCEGM) were manually contoured. Unsupervised pattern recognition was used to decompose the DCE data in prostate as the product W.H of k weight maps W and k patterns H, k=3. A well-perfused pattern (quick wash-in) Hwp was identified along with its weight map Wwp. Voxels with weight Wwp>60% determined a well-perfused volume ROIwp. Six quantitative features F were extracted from SROI=mean(ROIwp) and normalized by SGM =mean(DCEGM). Map of aggressiveness for feature F was constructed as F-map = (F(SROI, SGM)/mean(Wwp(ROIwp)))*Wwp and validated using 30 positive biopsy targets from 15 patients with MR-US guided biopsies. The relationship between features and Gleason Score (GS) was investigated using Spearman correlation coefficient and Kruskal-Wallis test.
Results: For 39 patients with PZ lesion, the volume of ROIwp and GS were significantly correlated. Three features were found statistically significant: late_AUC_ratio (ρ=0.32,p=0.049), early_AUFC_ratio (ρ=0.34,p=0.036) and wash-out_ratio (ρ=0.42,p=0.008), and three features were marginally significant: early_AUC_ratio (ρ=0.28,p=0.090), wash-in_ratio (ρ =0.28,p=0.090) and late_AUFC_ratio (ρ=0.29,p=0.074). For 30 targeted biopsies four features F were significantly correlated with GS: early_AUC_ratio (ρ= 0.57,p<0.001), wash-in_ratio (ρ=0.50,p=0.002), early_AUFC_ratio (ρ= 0.56,p=0.001), and wash-out ratio (ρ=0.34,p=0.045). AUC for indolent/aggressive for these features were 0.92 for early_AUC_ratio (p=0.018), 0.85 for wash-in_ratio (p=0.053), 0.95 for early_AUFC_ratio (p=0.012), and 0.88 for wash-out_ratio (p=0.032). F-maps were validated plotting F and F-map values for targeted biopsies: R2 correlation coefficients between F and F-map were larger than 0.85 for all features. Regression slope values were between 0.90 and 1.12.
Conclusion: Quantitative features from DCE-MRI, normalized by muscle are able to assess aggressiveness and map the areas of risk in the prostate.
Contact Email: