Multiphase CT-based prediction of Child-Pugh classification: a machine learning approach

Area of Science:

• Radiology
• Medical Imaging
• Machine Learning in Medicine

Background:

• Child-Pugh classification is crucial for assessing liver disease severity.
• Accurate prediction of Child-Pugh class is essential for patient management.
• Current methods rely on clinical and laboratory parameters.

Purpose of the Study:

• To evaluate machine learning algorithms for predicting Child-Pugh classification.
• To assess the feasibility of using multiphase computed tomography (CT) for this prediction.
• To compare the performance of machine learning models against experienced radiologists.

Main Methods:

• Retrospective study of 259 patients undergoing multiphase abdominal CT.
• Child-Pugh scores determined from clinical and laboratory data.
• Prediction models included Linear Regression (LR), Random Forest (RF), and Convolutional Neural Network (CNN); performance compared to Experienced Radiologists (ERs).

Main Results:

• Eleven imaging features significantly correlated with Child-Pugh class.
• CNN (ρ=0.51) and ERs (ρ=0.60) showed significant correlations in predicting Child-Pugh class.
• CNN and ERs demonstrated significantly better accuracies and AUCs for binary severity classification compared to LR and RF.

Conclusions:

• A Convolutional Neural Network (CNN) can effectively predict Child-Pugh class using multiphase CT.
• The performance of the CNN model is comparable to that of experienced radiologists.
• Machine learning offers a promising tool for objective liver disease severity assessment from imaging.