A Hybrid Machine Learning Model Based on Semantic Information Can Optimize Treatment Decision for Naïve Single 3-5-cm...

Area of Science:

• Hepatocellular Carcinoma (HCC) Research
• Machine Learning in Oncology
• Treatment Optimization Strategies

Background:

• Tumor recurrence remains a significant challenge for hepatocellular carcinoma (HCC) patients undergoing local treatment.
• Identifying optimal initial treatment strategies is crucial for improving patient outcomes.

Purpose of the Study:

• To develop a hybrid machine learning model for recommending the optimal first treatment (laparoscopic hepatectomy [LH] or microwave ablation [MWA]) for single 3-5-cm HCC patients.
• To predict the probability of early recurrence (ER; ≤2 years) to guide treatment decisions.

Main Methods:

• Retrospective analysis of 582 HCC patients (LH: 300, MWA: 282) with 20 semantic variables and ≥24 months follow-up.
• Building and comparing five machine learning algorithms (logistic regression, random forest, neural network, stochastic gradient boosting, XGBoost) to create optimized LH and MWA models.
• Developing a hybrid model integrating LH-XGBoost and MWA-stochastic gradient models to predict ER probability and guide treatment decisions.

Main Results:

• The hybrid model achieved high predictive accuracy, with LH-XGBoost (AUC=0.744) and MWA-stochastic gradient (AUC=0.750) models selected.
• Predicted ER probabilities closely matched actual ER rates (p > 0.05).
• Adherence to model recommendations significantly reduced ER rates (LH: 21.2% vs. 46.2%; MWA: 26.3% vs. 54.1%; p < 0.05), decreasing overall ER from 38.2% to 25.6% (p < 0.001).

Conclusions:

• The hybrid machine learning model accurately predicts ER probability for different treatments in single 3-5-cm HCC.
• This model provides reliable evidence for optimizing initial treatment decisions, thereby reducing tumor recurrence.
• The findings support the use of AI-driven tools for personalized HCC treatment planning.