Deep Learning Algorithm Based on Contrast-Enhanced Ultrasound Potentially Optimizes Treatment Strategies for Solitary...

Area of Science:

• Medical Imaging and Artificial Intelligence
• Oncology
• Surgical Oncology

Background:

• Hepatocellular carcinoma (HCC) recurrence after treatment poses a significant clinical challenge.
• Accurate prognostication is crucial for guiding treatment decisions in HCC patients.
• Current methods for predicting recurrence may not fully leverage advanced imaging and deep learning capabilities.

Purpose of the Study:

• To develop and validate a deep learning (DL) prognostic framework utilizing a dual-branch Vision Transformer (ViT) architecture.
• To stratify recurrence risk in solitary primary HCC patients undergoing surgical resection (SR) or thermal ablation (TA).
• To explore the framework's potential as a tool for aiding therapy-related discussions and treatment optimization.

Main Methods:

• A two-center study included 575 HCC patients with preoperative contrast-enhanced ultrasound (CEUS) images.
• Dual-branch ViT models (ViT-SR and ViT-TA) were developed by combining ViT-extracted CEUS features with clinical data for recurrence-free survival (RFS) stratification.
• Risk reclassification was performed by applying patients to alternative treatment models to investigate treatment optimization potential.

Main Results:

• The ViT-SR and ViT-TA models demonstrated favorable performance in validation cohorts, with C-Indexes of 0.73 and 0.76, respectively.
• High AUC, precision, and F1 scores were achieved, indicating robust predictive capabilities.
• Time-dependent AUC values remained stable, and satisfactory calibration was observed. Treatment strategy conversion analysis suggested potential benefits for a subset of patients under alternative treatments.

Conclusions:

• ViT-based DL models integrating multi-phase CEUS images and clinical variables enable non-invasive prediction of HCC recurrence.
• These models can assist in personalizing clinical treatment strategies by simulating treatment conversion.
• The framework offers a promising tool for enhancing therapeutic decision-making in HCC management.