Magnetic Resonance Deep Learning Radiomic Model Based on Distinct Metastatic Vascular Patterns for Evaluating...

Area of Science:

• Hepatobiliary Malignancies
• Medical Imaging
• Machine Learning in Oncology

Background:

• Hepatocellular carcinoma (HCC) metastasis involves microvascular invasion (MVI) and vessels encapsulating tumor clusters (VETC).
• Existing radiological studies primarily focus on predicting VETC status, with less emphasis on its combined prognostic value with MVI.

Purpose of the Study:

• To develop and compare predictive models for recurrence-free survival (RFS) in HCC patients.
• To evaluate clinical, radiomics, and deep learning models incorporating VETC and MVI for RFS prediction.

Main Methods:

• Retrospective analysis of 398 HCC patients undergoing resection, divided into training (n=358) and testing (n=40) cohorts.
• Development of clinical, radiomics, and deep learning models using MRI data (T1WI, T2WI, contrast-enhanced phases) and clinical factors (VETC, MVI, Barcelona stage, tumor diameter, AFP).
• Statistical analysis included COX regression, LASSO, Kaplan-Meier curves, and C-index for model performance evaluation.

Main Results:

• The deep learning model achieved the highest predictive performance with a C-index of 0.830 in the test cohort.
• Performance comparison showed the deep learning model outperformed the clinical-radiomic nomogram (0.731), radiomic signature (0.707), and clinical model (0.702).
• The average RFS for all patients was 26.77 months.

Conclusions:

• A magnetic resonance (MR) deep learning model integrating VETC and MVI offers a valuable tool for assessing survival outcomes in HCC patients.
• This approach has the potential to improve prognostic accuracy and guide clinical decision-making for HCC management.