CMT-FFNet: A CMT-based feature-fusion network for predicting TACE treatment response in hepatocellular carcinoma

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Hepatocellular carcinoma (HCC) is a primary liver cancer.
• Transarterial chemoembolization (TACE) is a common HCC treatment.
• Accurate prediction of TACE response is essential for personalized HCC management.

Purpose of the Study:

• To develop a novel deep learning model for predicting TACE treatment response in HCC.
• To leverage multiphase Magnetic Resonance Imaging (MRI) for enhanced prediction accuracy.

Main Methods:

• A feature fusion network (CMT-FFNet) based on Convolutional Neural Networks Meet Vision Transformers (CMT) was proposed.
• The model integrates local and global feature extraction using attention mechanisms.
• Orthogonality loss was introduced to optimize the fusion of imaging and clinical data.

Main Results:

• The CMT-FFNet effectively captured correlations within multiphase MRI data and multimodal inputs.
• The model demonstrated significantly improved prediction performance for TACE treatment response.
• Visualization techniques identified key regions influencing model predictions.

Conclusions:

• The proposed CMT-FFNet model shows high potential for accurately predicting TACE efficacy in HCC patients.
• Multiphase MRI combined with advanced AI offers a promising approach for preoperative treatment response prediction.
• This method can aid in individualized treatment strategies for HCC.