CT-Based Prediction of Liver Function and Post-PVE Hypertrophy Using an Artificial Neural Network

Area of Science:

• Radiology
• Artificial Intelligence
• Hepatology

Background:

• Portal vein embolization (PVE) is used to induce liver hypertrophy before major liver resection.
• Accurate prediction of hypertrophy and liver function is crucial for patient selection and surgical planning.
• Computed tomography (CT) texture analysis offers potential for non-invasive assessment of liver tissue characteristics.

Purpose of the Study:

• To evaluate the predictability of liver hypertrophy and maximum liver function capacity (LiMAx) using an artificial neural network (ANN) model.
• To assess the association between CT texture features and post-PVE liver hypertrophy and LiMAx.
• To determine the accuracy of an ANN model in predicting these outcomes.

Main Methods:

• Retrospective analysis of 118 patients undergoing CT before and after PVE for liver resection.
• LiMAx testing performed in 55 patients pre-PVE.
• A multilayer perceptron ANN model utilized to analyze CT texture features and predict hypertrophy and LiMAx.

Main Results:

• Median hypertrophy of 33.9% observed post-PVE; 36.5% of patients showed non-response (<25% hypertrophy).
• ANN model achieved 74.6% accuracy in predicting hypertrophy response (sensitivity 95.8%, specificity 44.4%).
• ANN model showed strong correlation with observed LiMAx (R² = 0.89).

Conclusions:

• ANN models utilizing CT texture features can effectively predict maximum liver function capacity (LiMAx).
• These models show promise in assessing potential liver hypertrophy following PVE.
• CT-based ANN analysis may enhance preoperative evaluation for extended liver resection.