An Automated Multi-Objective Model with Hyperparameter Optimization for Treatment Outcome Prediction in Metastatic Melanoma
Z Zhou1, M Zhou2, Z Wang3, X Chen2*, (1) University of Central Missouri, Warrensburg, MISSOURI, (2) Xi'an Jiaotong University,Xi'an ,Shaanxi,China, (3) Peking University Cancer Hospital & Institute, Beijing,China
Presentations
WE-F-TRACK 6-2 (Wednesday, 7/28/2021) 4:30 PM - 5:30 PM [Eastern Time (GMT-4)]
Purpose: Accurately predicting immunotherapy response is of great importance to improve treatment effectiveness in metastatic melanoma. We aim to develop a new automated multi-objective model with hyperparameter optimization (AutoMO-HO) for treatment outcome prediction.
Methods: Totally 50 patients who exam contrast-enhanced computed tomography (CECT) are used in this study. The images with both pre- and one-cycle post-treatment are obtained for extracting delta-radiomic features by calculating the difference between pre-treatment and one-cycle post-treatment radiomic features. To obtain more discriminative and complementary features, one-cycle post-treatment radiomic features are combined with delta-radiomic features to predict treatment response. To obtain the balanced sensitivity and specificity as well as higher confidence output, an automated multi-objective model (AutoMO) is applied in this study. However, there are several hyperparameters to be set manually before training, leading to the non-optimal model performance. As such, Bayesian optimization is introduced to automatically find the optimal model hyperparameter, and a new model termed as AutoMO-HO is developed based on AutoMO in this study. In AutoMO-HO, the training stage consists of two phases, they are Bayesian hyperparameter optimization through the Tree Parzen estimator algorithm and Pareto-optimal model set generation. In testing stage, the evidential reasoning (ER) strategy is used to fuse the output of each Pareto-optimal model to obtain more reliable results. Finally, the label with the maximal output confidence is taken as final output label. Accuracy, AUC, sensitivity and specificity are used for evaluation.
Results: In this study, five-fold cross validation is performed. Accuracy, AUC, sensitivity and specificity of AutoMO-HO can achieve 0.88, 0.91, 0.96 and 0.88, respectively, while AutoMO without hyperparameter optimization is 0.84, 0.87, 0.82 and 0.85.
Conclusion: A new automated multi-objective model with Bayesian hyperparameter optimization (AutoMO-HO) for predicting immunotherapy response is developed in metastatic melanoma. The experimental results demonstrated that AutoMO-HO can obtain better performance than AutoMO.
Handouts
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email