MRI-Based Topology Deep Learning Model for Noninvasive Prediction of Microvascular Invasion and Assisting Prognostic...

Area of Science:

• Hepatocellular Carcinoma Research
• Medical Imaging Analysis
• Deep Learning Applications

Background:

• Microvascular invasion (MVI) is a critical prognostic factor in hepatocellular carcinoma (HCC), negatively impacting patient outcomes.
• Current prediction methods for MVI often lack precision, necessitating improved preoperative assessment tools.
• Deep learning (DL) and topological analysis show promise in enhancing the predictive capabilities of medical imaging.

Purpose of the Study:

• To develop and externally validate a novel MRI-based deep learning (DL) model incorporating topological features for the preoperative prediction of MVI in HCC.
• To assess the model's performance in predicting MVI and its correlation with patient survival outcomes.

Main Methods:

• A retrospective study involving 589 surgically treated HCC patients from two centers.
• Development of a topology-CNN (TopoCNN) model and a TopoCNN+Clinical (TopoCNN+Clinic) model using automatic liver and tumor segmentation via DL.
• External validation of models using receiver operating characteristic curve analysis (AUC) and Cox regression for recurrence-free survival (RFS) and overall survival (OS).

Main Results:

• The TopoCNN and TopoCNN+Clinic models achieved high AUCs (0.871-0.895) in external validation for MVI prediction.
• For smaller tumors (≤3.0 cm), the TopoCNN+Clinic model showed strong performance (AUC 0.929 internally, 0.758 externally).
• TopoCNN-derived MVI prediction probability independently predicted early RFS (HR 6.64) and OS (HR 13.33).

Conclusions:

• An MRI topological DL model, leveraging automatic segmentation, accurately predicts MVI in HCC.
• The model effectively stratifies postoperative early recurrence-free survival and overall survival.
• This approach can significantly assist in personalized treatment decision-making for HCC patients.