Room: AAPM ePoster Library
Purpose: Prostate cancer is the second most common cancer among males and the fifth deadliest worldwide. Statin, an anti-cholesterolemic drug, and omega-3, a fatty acid supplement, were each recently found linked with the prostate cancer risk and aggressiveness. The observed associations are complex and controversial, drawing active research for further elucidation. We therefore explore the novel application of using radiomics to predict statin and omega-3 use in prostate cancer patients.
Methods: The T2-weighted MRI of 91 prostate cancer patients acquired at diagnosis were used for the study. Statin use and omega-3 use were collected from medical records. Two ROIs, the prostate and the peripheral zone (PZ), were manually segmented. From each ROI, 944 radiomic features (original, LOG and wavelet) were extracted using SlicerRadiomics after field bias correction and normalization. Heatmaps were generated to study the feature patterns. For each ROI/drug combination, features were selected using a univariate ANOVA test, a recursive correlation pruning step, and a sequential floating forward method, with a 1000-time 2/3 subsampling. Radiomic models were trained and evaluated with 500-round 3-fold cross-validation.
Results: Among the 91 patients, 42 had statin use and 28 had omega-3 use. The prostate ROI radiomics had weaker associations with these drug uses compared with the PZ ROI. For statin use, a 6-feature PZ-based radiomic signature achieved a mean AUC of 0.81 and AUPRC of 0.78 in cross-validation. For omega-3, a 4-feature PZ-based radiomic signature achieved a mean AUC of 0.73 and AUPRC of 0.64.
Conclusion: As the first study to analyze the radiomic feature pattern in relation to statin and omega-3 drug uses in prostate cancer patients, our study illustrated the potential usefulness of the radiomics tool for further exploring these drugs’ effects and mechanisms in prostate cancer.
Prostate Therapy, MRI, Quantitative Imaging
IM/TH- Image Analysis (Single Modality or Multi-Modality): Imaging biomarkers and radiomics