Deep Learning-Based Automatic Segmentation Combined with Radiomics to Predict Post-TACE Liver Failure in HCC Patients
Shuai Li1, Kaicai Liu1, Chang Rong1
1Department of Radiology, the First Affiliated Hospital of AnHui Medical University, Hefei, Anhui Province, People's Republic of China.
Journal of Hepatocellular Carcinoma
|December 23, 2024
Summary
A new deep learning model accurately predicts post-transarterial chemoembolization liver failure in hepatocellular carcinoma patients. This combined approach offers a valuable tool for treatment planning.
Area of Science:
- Oncology
- Medical Imaging
- Artificial Intelligence
Background:
- Hepatocellular carcinoma (HCC) patients undergoing transarterial chemoembolization (TACE) face risks of post-TACE liver failure (PTLF).
- Accurate prediction of PTLF is crucial for optimizing treatment strategies and patient management.
Purpose of the Study:
- To develop and validate a deep learning-based automatic segmentation model for medical images.
- To integrate radiomics features with clinical data for predicting PTLF in HCC patients.
Main Methods:
- A retrospective study of 210 TACE-treated HCC patients.
- Development of an nnU-Net based automatic segmentation model, evaluated using Dice Similarity Coefficient (DSC).
- Creation of combined, clinical, and radiomics predictive models using logistic regression and assessed via AUC, calibration curves, and DCA.
Main Results:
- The automatic segmentation model achieved high performance (DSC: 83.05% for tumor, 92.72% for non-tumoral parenchyma).
- International Normalized Ratio (INR) and Albumin (ALB) were identified as independent clinical predictors.
- The combined model demonstrated superior predictive performance (AUC: 0.878) compared to clinical (AUC: 0.785) and radiomics (AUC: 0.815) models.
Conclusions:
- A reliable combined predictive model integrating deep learning segmentation and radiomics can accurately predict PTLF in HCC patients.
- This model serves as a valuable reference for clinicians in formulating treatment plans.
