Contrast-Enhanced CT Shell Features and Deep Learning for Predicting Early Transarterial Chemoembolization...

Area of Science:

• Radiology
• Artificial Intelligence in Medicine
• Oncology

Background:

• Hepatocellular carcinoma (HCC) management often involves transarterial chemoembolization (TACE).
• Predicting early TACE refractoriness (ETR) is crucial for timely treatment modification.
• Current prediction methods may lack precision in identifying patients unlikely to respond to initial TACE.

Purpose of the Study:

• To develop and validate a predictive model for early TACE refractoriness (ETR) in hepatocellular carcinoma (HCC) patients.
• To integrate contrast-enhanced CT (CECT) shell features with the Vision-Mamba (Vim) architecture for enhanced prediction accuracy.
• To compare the performance of the proposed Vim-based model against traditional machine learning models.

Main Methods:

• A retrospective, two-center study involving 447 HCC patients (254 training, 108 validation, 75 testing).
• Development of a joint predictive model using Vim architecture and CECT shell features.
• Performance evaluation using accuracy, AUC, calibration curves, DCA, and SHAP analysis.

Main Results:

• The combined Vim-based model demonstrated superior performance with high AUC values (0.959 training, 0.956 validation, 0.942 testing).
• Calibration curves and decision curve analysis confirmed the model's clinical practicality.
• SHAP analysis provided visual interpretability of the model's predictions.

Conclusions:

• The Vim-based model integrating CECT and shell features shows significant promise for predicting ETR in HCC patients.
• This model can serve as a preliminary tool for patient stratification.
• Further prospective validation is necessary due to the retrospective nature of the study.