5455. Validation of a Nomogram Based on Volumetric MR Radiomics and the Okuda System in Predicting Overall Survival in Patients Undergoing TACE
Authors* Denotes Presenting Author
Mohammad MirzaAghazadeh Attari *;
Johns Hopkins Health Institutions
Ghazal Zandieh ;
Johns Hopkins Health Institutions
Shadi Afyouni ;
Johns Hopkins Health Institutions
Ali Borhani ;
Johns Hopkins Health Institutions
Alireza Mohseni ;
Johns Hopkins Health Institutions
Iman Yazdani Nia ;
Johns Hopkins Health Institutions
Ihab Kamel ;
Johns Hopkins Health Institutions ; University of Colorado-Denver, Anschutz Medical Campus
Objective:
To develop a nomogram model, which is a graphical representation of a regression model that incorporates quantitative imaging markers, semantic imaging signs and established clinical criteria.
Materials and Methods:
This study was performed using retrospective imaging and clinical data gathered from 193 (153 patients used for development and 40 used for testing) patients with HCC undergoing TACE. MRI was preformed prior to initiation of therapy. All images were rescaled and normalized. All lesions were segmented manually, and 845 radiomics features were extracted using Pyradiomics 3.1. Linear least absolute shrinkage and selection operator was used to determine the relevant features, and a radiomics score was generated based on their respective coefficients. Tumor viability was determined using a previously determined cut-off based on MR-pathology correlation. A high-dimensional Cox model object (model type: minimax concave penalty) was used to generate a nomogram, and three risk groups. Log rank test was used to compare survival among these three groups. The model was validated and calibrated using the external dataset. All statistical analysis was performed using R (hdnom_5.0 package).
Results:
The mean age of the population cohort was 65 ± 12 years. There were no significant differences in regards to age, sex, and survival time in the test and training cohorts. After multiple-regression was performed on clinical and radiomics variables, a nomogram consisting of the OKUDA system, radiomics score, viable tumor burden (determined based on a predetermined threshold), and tumor margin was constructed and validated using external data. The time-AUC curves of the test and train cohorts were consistently above 0.75 Calibration curves showed high degrees of consistency among expected and predicted survival values, as well as Kaplan Meier curves of the training and testing cohorts. There was a significant difference in survival (log rank test) in different risk groups determined by the model (<em>p</em> = 0.009 and 0.0001 in the train and testing datasets, respectively).
Conclusion:
Incorporation of radiomics features, multiparametric MR features and clinical criteria enables differentiation between responders and nonresponders to TACE therapy and can be used in predicting overall survival in patients with HCC undergoing treatment. Concomitant utilization of quantitative imaging data may help in avoiding unnecessary therapy or determining patients who may be in need of multiple treatment sessions.
