Abstract PO-059: Recurrence Predictive Models for Patients with Hepatocellular Carcinoma after Radiofrequency Ablation Based on Machine Learning Algorithms

Abstract Background: Hepatocellular carcinoma (HCC) has been one of the leading causes of cancer death. Despite accurate diagnosis and effective treatments, the recurrence rate of HCC is still high. To build a recursive model by analyzing preoperative reports is helpful for follow-up and observing recurrence of tumor. Methods: Newly diagnosed HCC patients in National Taiwan University Hospital (NTUH) who received radiofrequency ablation (RFA) as the first treatment, were enrolled. Among 334 enrolled patients, 256 patients did not have recurrent HCC one year after RFA treatment and the other 78 patients had HCC recurrences. Data was processed with workflow of feature extraction and data imputation to acquire 16 features. We use different machine-learning methods to build the recurrence prediction model of HCC. The model performance among linear regression (LR), support vector machine (SVM), random forest (RF), and deep neural network (DNN) under a variety of experimental environment, including different parameters of models, data normalizations, and methods of data up-sampling were compared. Results: DNN model has the relatively best result that is the highest accuracy (82.65%) and second balanced accuracy (BAC=66.03%) among all models. SVM with feature selection according to importance in RF has the best BAC result (66.48%) with 10% lower accuracy (72.51%) compared with DNN. LR has the worst BAC (50.00%). Conclusion: Among all prediction models, DNN achieved the best results by considering both accuracy and balanced accuracy. Establishment of prediction model could identify the risk patients with close monitoring of tumor recurrence. Citation Format: Ja-Der Liang, Ta-Wei Yang, Po-When Chen, Cheng-Fu Chou, Yao-Ming Wu. Recurrence predictive models for patients with hepatocellular carcinoma after radiofrequency ablation based on machine learning algorithms [abstract]. In: Proceedings of the AACR Virtual Special Conference on Artificial Intelligence, Diagnosis, and Imaging; 2021 Jan 13-14. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(5_Suppl):Abstract nr PO-059.