Preoperative CT-based Radiomics for Predicting Response to Neoadjuvant Chemoimmunotherapy in Esophageal Squamous Cell...

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Esophageal squamous cell carcinoma (ESCC) requires effective prediction of neoadjuvant chemoimmunotherapy response.
• Current prediction methods may lack accuracy and comprehensive feature integration.

Purpose of the Study:

• To evaluate a CT-based model integrating 2D and 2.5D deep learning (DL) with radiomic features for predicting neoadjuvant chemoimmunotherapy response in ESCC patients.

Main Methods:

• Retrospective study with training, internal validation, and external testing cohorts from multiple cancer centers.
• Radiomic features extracted manually; 2D and 2.5D deep transfer learning (DTL) features derived from pretrained DL networks.
• Support vector machine (SVM) model utilized for optimal performance assessment based on AUC.

Main Results:

• A fusion model combining SVM with ResNet18-based DTL features demonstrated superior performance.
• Achieved AUCs of 0.85 for 2D DTL and 0.84 for 2.5D DTL in the external testing group.
• The integrated model effectively predicted treatment response.

Conclusions:

• A fusion model integrating 2D/2.5D DTL and radiomic features is effective for predicting neoadjuvant chemoimmunotherapy response in ESCC.
• This artificial intelligence approach enhances prognosis and prediction capabilities.
• CT-based deep learning holds significant potential for personalized cancer treatment strategies.