P359 Radiomics could predict surgery at 10 years in Crohn’s disease
Laterza, L.(1);Boldrini, L.(2);Tran, H.E.(2);Votta, C.(2);Larosa, L.(2);Minordi, L.M.(2);Scaldaferri, F.(1);Papa, A.(1);Armuzzi, A.(1);Gasbarrini, A.(1);
(1)Fondazione Policlinico Universitario A. Gemelli IRCCS- Università Cattolica del Sacro Cuore Roma, CEMAD- Digestive Diseases Center, Rome, Italy;(2)Fondazione Policlinico Universitario A. Gemelli IRCCS- Università Cattolica del Sacro Cuore Roma, Department of Diagnostic Imaging- Oncological Radiotherapy- and Hematology – Diagnostic Imaging Area, Rome, Italy;
Background
Predicting clinical outcomes and defining the best time for elective surgery represents a major challenge in managing Crohn’s disease (CD) patients. Radiomics provides a method to extract quantitative features from medical images and may successfully predict clinical course. Although radiomics has been successfully used in oncology, its application in inflammatory bowel disease is still scarce. The aim of this pilot study is to evaluate the use of radiomics to predict the need for surgery in a cohort of CD patients with a long-term follow-up.
Methods
We retrospectively selected a cohort of 30 CD patients undergone one or more CT scan enterographies between 2009 and 2011 for a total of 44 CT scans. For each CT scan, typical lesions of CD were searched by an expert radiologist generating a region of interest (ROI) segmentation for each lesion. 93 lesions were lastly obtained for radiomic analysis. 217 radiomic features were extracted from each ROI using a dedicated institutional R library (Moddicom) validated within the Image Biomarker Standardization Initiative. Patients’ charts were reviewed to assess if patients underwent surgery in a 10-year follow-up for a binary classification.
The feature selection process included a univariate analysis with the Wilcoxon-Mann-Whitney test, the application of Boruta algorithm and the exclusion of highly correlated features. A logistic regression model was built with the selected features and evaluated by computing the area under the curve (AUC) of the receiver operating characteristic (ROC) curve. Accuracy, sensitivity, specificity, positive and negative predictive values (PPV, NPV) of the model were calculated using the best cut-off according to the Youden index method. A 3-fold cross-validation repeated 5 times was used for internal validation.
Results
Two radiomic features were obtained from the feature selection process and resulted statistically significant in the fitted logistic regression model (p < 0.0001) in predicting surgery: grey level histogram variance and grey level non-uniformity. This model presented an AUC (95% CI) of 0.83 (0.73-0-91) in predicting surgery, confirmed by the cross-validation. Mean (standard deviation) values of the model performance metrics over the cross-validation iterations were: accuracy 0.78 (0.02), sensitivity 0.68 (0.14), specificity 0.86 (0.07), PPV 0.72 (0.12), NPV 0.83 (0.09).
Conclusion
Radiomics could be a helpful tool to identify patients with high risk for surgery and needing a stricter monitoring or therapy intensification. Further studies are required to understand the possible corresponding histopathological features and to obtain larger and external validations of this model.