A Machine Learning Guided System for Therapeutic Decision-Making in Intra-Arterial Therapy of Hepatocellular
Linling Tan1, Rongshan Fan1, Jun Zhang1
1Department of Hepatology, Shenzhen Hospital of Integrated Traditional Chinese and Western Medicine, Shenzhen, Guangdong, People's Republic of China.
Objective:
To develop and validate a machine learning-guided system (MLGS) for stratifying prognostic risk of unresectable hepatocellular carcinoma (uHCC) based on clinical data before and after intra-arterial therapy (IAT).
Methods:
Between April 2008 and June 2022, a total of 5646 eligible patients with uHCC who underwent initial IAT were consecutively identified in 15 hospitals. The 5 year mortality was used as primary predictive outcome. Thirty-five sets of clinical data, including 29 preoperative and 6 postoperative data, were input successively into five supervised ML models. The performance of the ML models was compared using the area under the receiver operating characteristic (AUC) curve with the DeLong test. Kaplan-Meier analysis was used to revealed overall survival (OS) risk stratification of the MLGS.
Results:
These patients were divided into training datasets (TD, n=3387), internal testing datasets (ITD, n=1130), and external test datasets (ETD, n=1129), respectively.The CatBoost model yield the best discrimination using preoperative data in the ML models using 23 variables, the AUC of CatBoost model were 0.777-0.735 in three datasets. Meanwhile, the XGBoost model yield the best discrimination using postoperative 20 variables and the AUC of XGBoost model were 0.904-0.861 in three datasets, which outperform significantly the performance of the CatBoost model (DeLong test, P < 0.001). The MLGS based on the XGBoost model provide significantly different OS between three risk stratification in three datasets (all, P < 0.001).
Conclusion:
The MLGS may guide radiologists in developing strategies of IAT for uHCC. Prospective studies are needed to evaluate its clinical utility.
