CT-based deep learning predicts immunotherapy response in esophageal squamous cell cancer patients.
Authors
Affiliations (4)
Affiliations (4)
- Department of Digital Medicine, College of Biomedical Engineering and Medical Imaging, Army Medical University (Third Military Medical University), Chongqing, China.
- Yu-Yue Pathology Research Center, Jinfeng Laboratory, Chongqing, China.
- Chongqing Engineering Research Center of Medical Electronics and Information Technology, Chongqing University of Posts and Telecommunications, Chongqing, China.
- Department of Thoracic Surgery, Southwest Hospital/The First Affiliated Hospital of Army Medical University, Army Medical University (Third Military Medical University), Chongqing, China.
Abstract
Immunotherapy has emerged as a promising addition to neoadjuvant therapy for locally advanced esophageal squamous cell carcinoma (LA-ESCC), yet treatment response varies substantially among patients. Reliable biomarkers for predicting response to immunochemotherapy (ICT) remain lacking. We aimed to develop and externally validate a deep learning model for predicting immune Response Evaluation Criteria in Solid Tumors (iRECIST) response to ICT in LA-ESCC. This retrospective, dual-center study included 290 patients with pathologically confirmed LA-ESCC. Of these, 264 patients from The First Affiliated Hospital of Army Medical University were randomly divided into training (n=185) and internal testing cohorts (n=79), and 26 patients from Banan District People's Hospital of Chongqing served as an external testing cohort. A deep learning-based model was developed using pretreatment contrast-enhanced computed tomography (CT) images to predict short-term iRECIST-defined therapeutic response. Model discrimination, calibration, and clinical utility were assessed using the area under the receiver operating characteristic curve (AUC), bootstrap 95% confidence intervals (CIs), Brier scores, calibration intercept and slope, and decision curve analysis (DCA). The model achieved AUCs of 0.853 (95% CI: 0.795-0.906), 0.794 (95% CI: 0.683-0.886), and 0.759 (95% CI: 0.549-0.925) in the training, internal testing, and external testing cohorts, respectively. Calibration was acceptable in the training and internal testing cohorts but weakened in the small external cohort (slope =0.36; 95% CI: -0.02 to 0.75), and DCA showed greater net benefit than "treat-all" and "treat-none" strategies across a range of threshold probabilities, with wider variability in the external cohort. In this dual-center retrospective cohort, a deep-learning model based on pretreatment contrast-enhanced CT showed exploratory predictive value for short-term radiological response to ICT in LA-ESCC. Given the small external cohort and single-region design, the present results should be regarded as hypothesis-generating, and prospective multi-regional validation is required before any clinical use.