MP18-06: Multi-Institutional Development and Validation of a Radiomic Model to Predict Prostate Cancer Recurrence following Radical Prostatectomy
Friday, May 3, 2024 3:30 PM to 5:30 PM · 2 hr. (US/Central)
302B
Abstract
Information
Full Abstract and Figures
Author Block
Linda M Huynh*, Benjamin Bonebrake, Omaha, NE, Joshua Tran, Oorange, CA, Jacob Marasco, Omaha, NE, Thomas E Ahlering, Oorange, CA, Shuo Wang, Michael Baine, Omaha, NE
Introduction
Multi-parametric magnetic resonance imaging (mpMRI)-derived radiomics have been shown to capture sub-visual patterns for quantitative characterization of prostate cancer (PC) phenotypes. The present study seeks to develop, test, and compare the performance of an MRI-derived radiomic model for the prediction of PC recurrence following definitive treatment with radical prostatectomy (RP).
Methods
mpMRI was obtained from 251 patients who had a minimum of 2 years follow-up following RP at two institutions. The prostate was manually delineated as the region of interest and 924 radiomic features were extracted. All features were tested for stability via intraclass correlation coefficient (ICC) and image normalization via histogram matching.
Results
Fourteen important and non-redundant features were found to be predictors of PC recurrence at a mean±SD of 3.2±2.2 years and were aggregated into a radiomic model. Five-fold, ten-run cross-validation yielded a receiver-operator characteristic area under the curve (ROC-AUC) of 0.89±0.04 in the training set (n=225). In comparison, the University of California San Fransisco Cancer of the Prostate Risk Assessment score (UCSF-CAPRA) and Memorial Sloan Kettering Cancer Center (MSKCC) Pre-Radical Prostatectomy nomograms yielded AUC of 0.66±0.05 and 0.67±0.05, respectively (p<0.01). Finally, when the radiomic model was applied to the test set (n=26), ROC-AUC was 0.78 and sensitivity, specificity, positive predictive value, and negative predictive value were 60%, 86%, 52% and 89%, respectively. Accuracy of the radiomic model in predicting PC recurrence was 81%.
Conclusions
The present study is a proof of concept for the use of an mpMRI-derived radiomic model in predicting PC recurrence in 251 prostate cancer patients, yielding fourteen radiomic features significantly associated with recurrence following RP. When these features were aggregated into a radiomic signature, this signature predicted recurrence well in cross-validation and predicted patients at risk for PC recurrence with 81% accuracy.
Source Of Funding
NA
Author Block
Linda M Huynh*, Benjamin Bonebrake, Omaha, NE, Joshua Tran, Oorange, CA, Jacob Marasco, Omaha, NE, Thomas E Ahlering, Oorange, CA, Shuo Wang, Michael Baine, Omaha, NE
Introduction
Multi-parametric magnetic resonance imaging (mpMRI)-derived radiomics have been shown to capture sub-visual patterns for quantitative characterization of prostate cancer (PC) phenotypes. The present study seeks to develop, test, and compare the performance of an MRI-derived radiomic model for the prediction of PC recurrence following definitive treatment with radical prostatectomy (RP).
Methods
mpMRI was obtained from 251 patients who had a minimum of 2 years follow-up following RP at two institutions. The prostate was manually delineated as the region of interest and 924 radiomic features were extracted. All features were tested for stability via intraclass correlation coefficient (ICC) and image normalization via histogram matching.
Results
Fourteen important and non-redundant features were found to be predictors of PC recurrence at a mean±SD of 3.2±2.2 years and were aggregated into a radiomic model. Five-fold, ten-run cross-validation yielded a receiver-operator characteristic area under the curve (ROC-AUC) of 0.89±0.04 in the training set (n=225). In comparison, the University of California San Fransisco Cancer of the Prostate Risk Assessment score (UCSF-CAPRA) and Memorial Sloan Kettering Cancer Center (MSKCC) Pre-Radical Prostatectomy nomograms yielded AUC of 0.66±0.05 and 0.67±0.05, respectively (p<0.01). Finally, when the radiomic model was applied to the test set (n=26), ROC-AUC was 0.78 and sensitivity, specificity, positive predictive value, and negative predictive value were 60%, 86%, 52% and 89%, respectively. Accuracy of the radiomic model in predicting PC recurrence was 81%.
Conclusions
The present study is a proof of concept for the use of an mpMRI-derived radiomic model in predicting PC recurrence in 251 prostate cancer patients, yielding fourteen radiomic features significantly associated with recurrence following RP. When these features were aggregated into a radiomic signature, this signature predicted recurrence well in cross-validation and predicted patients at risk for PC recurrence with 81% accuracy.
Source Of Funding
NA
Sessions
MP18: Imaging/Uroradiology I
302B