Purpose: To predict the clinical evolution of patients with non-clinically significant prostate cancer (PCa) scheduled for active surveillance. Materials and Methods: 3T-mpMRI examinations at baseline and follow-up were retrieved from our institutional database for 16 patients in active surveillance, equally split between stability and progression disease. All mpMRI sequences were analysed by two experienced radiologists reporting several clinical parameter for both examinations of each patient. Accordingly, patients were assigned to stability or progression classes. After outlining prostate in T2w-MRI and aligning it on DW-MRI, 84 radiomic features were extracted from DWI (b=2000 s/mm2) and their differences between baseline and follow-up were computed. In order to prevent overfitting, one-only feature was selected through LASSO for discriminating the two classes, based on the AUC of the ROC curves and Wilcoxon rank-sum and sign-rank tests (p<0.05). Results: The feature selected by LASSO was the entropy of local skewness (sL-e) that discriminates the two classes (p=0.022) with AUC=0.73, specificity=75%, sensitivity=75%. sL-e shows a non-significant increase in stable patients (p=0.055), while significantly increases in patients in progression (p=0.039). Conclusions: Results highlight the promising role of sL-e, a compound feature referring to local tissue heterogeneity, whose values markedly increase in PCa progression. This early finding emphasizes the promising role of DWI-based radiomics in monitoring active surveillance patients, also confirming the established relationship between tumour progression and the increase of the variability of the local cellularization. Limitations: This discrimination between patients in stability and progression is achieved on a small dataset. However, the radiomic feature identified is the best candidate for the subsequent classification step to carry out on a wider dataset to assess its predictive power.
Alessandro Bevilacqua, Fabio Ferroni, Margherita Mottola, Domenico Barone, Giampaolo Gavelli (2020). Predicting clinical evolution of prostate cancer in active surveillance: early suggestions from DW-MRI [10.26044/ECR2020/C-13608].
Predicting clinical evolution of prostate cancer in active surveillance: early suggestions from DW-MRI
Alessandro Bevilacqua;Margherita Mottola;Giampaolo Gavelli
2020
Abstract
Purpose: To predict the clinical evolution of patients with non-clinically significant prostate cancer (PCa) scheduled for active surveillance. Materials and Methods: 3T-mpMRI examinations at baseline and follow-up were retrieved from our institutional database for 16 patients in active surveillance, equally split between stability and progression disease. All mpMRI sequences were analysed by two experienced radiologists reporting several clinical parameter for both examinations of each patient. Accordingly, patients were assigned to stability or progression classes. After outlining prostate in T2w-MRI and aligning it on DW-MRI, 84 radiomic features were extracted from DWI (b=2000 s/mm2) and their differences between baseline and follow-up were computed. In order to prevent overfitting, one-only feature was selected through LASSO for discriminating the two classes, based on the AUC of the ROC curves and Wilcoxon rank-sum and sign-rank tests (p<0.05). Results: The feature selected by LASSO was the entropy of local skewness (sL-e) that discriminates the two classes (p=0.022) with AUC=0.73, specificity=75%, sensitivity=75%. sL-e shows a non-significant increase in stable patients (p=0.055), while significantly increases in patients in progression (p=0.039). Conclusions: Results highlight the promising role of sL-e, a compound feature referring to local tissue heterogeneity, whose values markedly increase in PCa progression. This early finding emphasizes the promising role of DWI-based radiomics in monitoring active surveillance patients, also confirming the established relationship between tumour progression and the increase of the variability of the local cellularization. Limitations: This discrimination between patients in stability and progression is achieved on a small dataset. However, the radiomic feature identified is the best candidate for the subsequent classification step to carry out on a wider dataset to assess its predictive power.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
