Multiparametric MRI-Based Machine Learning Radiomics Prognostic Models for Multifocal Hepatocellular Carcinoma Beyond...

Area of Science:

• Hepatobiliary Malignancies
• Medical Imaging and Radiomics
• Machine Learning in Oncology

Background:

• Multifocal hepatocellular carcinoma (MHCC) beyond Milan criteria presents a challenge for accurate preoperative risk stratification.
• Existing models may not fully capture the prognostic complexity of advanced MHCC.
• Novel approaches are needed to improve survival prediction in these patients.

Purpose of the Study:

• To develop and validate machine learning radiomics models for preoperative risk stratification of MHCC.
• To identify distinct radiomics-based patient subtypes for MHCC.
• To assess the added value of radiomics in predicting overall survival (OS) and recurrence-free survival (RFS) beyond conventional criteria.

Main Methods:

• Retrospective analysis of 156 MHCC patients beyond Milan criteria using multiparametric MRI (mpMRI).
• Extraction of radiomic features from tumor and peritumor regions.
• Application of spectral clustering for subtype identification and extreme gradient boosting (XGBoost)-LASSO Cox regression for risk score generation.

Main Results:

• Two distinct radiomics-based patient subtypes were identified, with one subtype showing significantly worse OS and RFS.
• Combined models incorporating radiomics predictors demonstrated improved OS prediction (C-index: 0.712 training, 0.710 validation) and RFS prediction (C-index: 0.735 training, 0.698 validation).
• The models achieved good predictive performance for 5-year OS (AUC: 0.77 training, 0.75 validation) and RFS (AUC: 0.81 training, 0.76 validation).

Conclusions:

• Machine learning radiomics models integrating mpMRI-derived features effectively predict outcomes in MHCC patients beyond Milan criteria.
• These models offer improved preoperative risk stratification compared to conventional methods.
• The findings support the use of radiomics for personalized treatment planning in advanced MHCC.