An interpretable machine learning model based on contrast-enhanced CT parameters for predicting treatment response to...

Area of Science:

• Radiology and Imaging
• Oncology
• Machine Learning in Medicine

Background:

• Intermediate-stage hepatocellular carcinoma (HCC) requires effective first-line therapies.
• Conventional transarterial chemoembolization (cTACE) is a standard treatment for intermediate-stage HCC.
• Predicting treatment response to cTACE is crucial for optimizing patient outcomes.

Purpose of the Study:

• To evaluate pre-therapy computed tomography (CT) parameters for predicting cTACE response in intermediate-stage HCC.
• To develop an interpretable machine learning model for treatment response prediction.
• To identify patients most likely to benefit from cTACE to avoid ineffective treatments.

Main Methods:

• Retrospective analysis of 367 intermediate-stage HCC patients treated with cTACE.
• Calculation of arterial (AER), portal venous (PER), and arterial portal venous (APR) enhancement ratios from multi-phase contrast-enhanced CT.
• Development and validation of machine learning models, including a Random Forest (RF) model, using CT parameters and clinical variables.
• Interpretation of the best-performing model using Shapley additive explanation (SHAP).

Main Results:

• Clinical predictors like tumor size and ECOG status were identified.
• Integrating CT parameters significantly improved prediction performance (net reclassification index = 0.318).
• The RF-combined model achieved an AUC of 0.800 on an external validation dataset and demonstrated significant survival stratification.

Conclusions:

• The developed RF-combined model is a robust and interpretable tool for predicting cTACE response in intermediate-stage HCC.
• This model can help select appropriate patients for cTACE, potentially sparing others from unnecessary or ineffective treatments.
• Improved patient selection can lead to better resource allocation and treatment efficacy.