ARRS 2022 Abstracts

RETURN TO ABSTRACT LISTING


E1709. Performance of Non-Gaussian Diffusion Models in mpMRI Differentiation of Peripheral Zone Prostate Cancer Grade
Authors
  1. Zhihua Li; University of Illinois at Chicago
  2. Guangyu Dan; University of Illinois at Chicago
  3. Vikram Tammana; University of Illinois at Chicago
  4. Scott Johnson; University of Illinois at Chicago
  5. Behnam Rabiee; University of Illinois at Chicago
  6. Joe Zhou; University of Illinois at Chicago
  7. Karen Xie; University of Illinois at Chicago
Objective:
The goal of this study is to compare the performance of ADC versus non-Gaussian diffusion models: diffusion kurtosis imaging (DKI) and fraction order calculus (FROC) in mpMRI differentiation of prostate cancer (PCa) tumor grade.

Materials and Methods:
Seventy-five patients who underwent targeted MRI-guided TRUS prostate biopsy within 6 months of MRI were reviewed retrospectively. ROIs were placed on suspicious lesions on MRI scans and then correlated to pathological results, i.e., Gleason scores (GS), based on core biopsy location. The final tumor count was 23 of GS 6 (3+3), 54 of GS 7, and 19 of GS 8. DWI scans of each lesion were fitted into the models to calculate ADC. The diffusion coefficient (Dapp for DKI, D for FROC) and kurtosis constant (Kapp for DKI), heterogeneity index (ß for FROC), and spatial constant (µ for FROC) were caluclated. The parameters of each model were combined via logistic regression then evaluated inROC analyses.

Results:
The group means of parameters from each model demonstrated significant differences (one-way ANOVA, p < 0.05) between each PCa grade. In differentiating National Comprehensive Cancer Network (NCCN) low risk from intermediate PCa (GS 6 vs 7), DKI (Dapp + Kapp; 0.824) and FROC (D + ß; 0.829) parameters had a larger AUC value compared to ADC (0.655). Between NCCN intermediate and high (GS 7 vs 8), FROC ß provided the greater AUC (0.719) compared to DKI (Dapp; 0.683) and ADC (0.595). In comparing clinically insignificant versus significant PCa (GS 6 vs 7), FROC (D + ß; 0.801) had a larger AUC value compared to DKI (Kapp; 0.776) and ADC (0.671).

Conclusion:
Non-gaussian diffusion models FROC and DKI demonstrate higher performance in differentiating PCa grades compared to standard ADC. FROC demonstrates higher performance than DKI in each PCa grade comparison, excluding the differentiation of GS 6 versus 7 lesions, where both models displayed similar performance.