Related Experiment Videos
Multivariable Radiomics Model for Predicting Programmed Death-Ligand 1 Expression After Neoadjuvant Chemoradiotherapy
Hyun Do Jung1,2, Tae Hoon Lee3, Byoung Hyuck Kim4,5
1Department of Artificial Intelligence, School of Computing, Yonsei University, Seoul, Republic of Korea.
Cancer Medicine
|August 10, 2026
Summary
This study developed models to predict programmed death-ligand 1 (PD-L1) expression in esophageal squamous cell carcinoma after neoadjuvant chemoradiotherapy. A combined clinical-radiomics approach showed feasible performance, though radiomics offered limited added value over clinical factors alone.
Area of Science:
- Oncology
- Radiology
- Machine Learning
Background:
- Esophageal squamous cell carcinoma (ESCC) requires predictive biomarkers for treatment response.
- Programmed death-ligand 1 (PD-L1) expression is crucial for immune checkpoint inhibitor efficacy.
- Neoadjuvant chemoradiotherapy (nCRT) is a standard treatment for locally advanced ESCC.
Purpose of the Study:
- To develop and evaluate models for predicting post-nCRT PD-L1 expression in ESCC.
- To integrate radiomics, deep learning, and machine learning for enhanced prediction.
- To assess the incremental value of radiomics over clinical factors.
Main Methods:
- Retrieved pre- and post-nCRT CT images and clinical data from 101 locally advanced ESCC patients.
- Segmented tumors for radiomic feature extraction and selection.
- Developed predictive models using logistic regression, machine learning (ResNet, TabNet), and clinical variables.
Main Results:
- Six radiomics features and key clinical factors (radiation technique, dose) were identified as predictors.
- The combined clinical-radiomics model achieved a weighted AUROC of 0.8214.
- Radiomics provided minimal incremental discrimination compared to clinical variables alone.
Conclusions:
- A combined clinical-radiomics model is feasible for assessing post-nCRT PD-L1 expression in ESCC.
- The incremental value of radiomics over clinical factors was limited.
- Findings are hypothesis-generating and require external validation for clinical application.