A Multi-Objective Radiomic Model for Predicting Survival in Patients with Oropharyngeal Cancer
X Pan1*, C Liu2, T Feng3, X Qi4, (1) Xi'an University of Posts and Telecommunications, Xi'an,shaanxi, CN, (2) Xi'an University of Posts and Telecommunications, Xi'an,shaanxi, CN, (3) Xi'an University of Posts and Telecommunications, Xi'an,shaanxi, CN, (4) UCLA School of Medicine, Los Angeles, CA
Presentations
PO-GePV-M-347 (Sunday, 7/10/2022) [Eastern Time (GMT-4)]
ePoster Forums
Purpose: Oropharyngeal cancer (OPC) is a malignant tumor with 5-years survival rate after radiotherapy less than 65%. Accurate predicting the survival treatment response of OPC patients allows for personalized treatment. We aim to develop a multi-objective radiomic model predicting 5-year survival for patients with OPC.
Methods: We proposed a multi-objective radiomic model to provide survival prediction for patients with OPC. The model Inputs include 34 clinical features and 1583 radiomic features extracted from CT images. In the model, we conduct feature selection using a multi-objective particle swarm optimization algorithm, where the number of features, specificity and sensitivity are considered as optimization objectives. To ensure that the selected features can better distinguish between false-positive and false-negative patients, we proposed an evaluation function based on the Gray Wolf Optimization SVM (GWO-SVM) algorithm to evaluate the particles. The model performance was evaluated using the area-under-ROC-curve (AUC), sensitivity and specificity on a cohort of 427 patients with OPC from TCIA database, where 341 cases were used for training; 86 cases were used as an independent cohort.
Results: The multi-objective approach achieved better performance than the model constructed using a single-objective approach. Incorporating GWO-SVM algorithm with the multi-objective approach, the results further improved compared with the multi-objective approach. The AUC, sensitivity, and specificity of the single-objective, multi-objective, and proposed methods were 72%, 90%, and 53%; 76%, 67%, and 57%; and 80%, 81%, and 63%, respectively.
Conclusion: The proposed GWO-SVM algorithm ensures the diversity of the selected particles, which translates to improved model performance. The proposed multi-objective radiomic model shows superiority in predicting the survival rate of OPC patients.
Funding Support, Disclosures, and Conflict of Interest: Supported by the National Natural Science Foundation of China(No.62001380); General Special Scientific Research Program of Shaanxi Provincial Education Department(20JK0910).
Keywords
CT, Quantitative Imaging, Feature Selection
Taxonomy
TH- Response Assessment: Radiomics/texture/feature-based response assessment
Contact Email