Multimodal deep learning radiomics model for predicting esophageal fistula after chemotherapy and/or radiotherapy in esophageal squamous cell carcinoma: a multicenter study

Objectives This study aims to develop a multimodal model that integrates deep learning radiomics and clinical characteristics to predict the risk of esophageal fistula (EF) after chemotherapy and/or radiotherapy in patients. Methods In this multicentre retrospective study, 186 patients with esophageal squamous cell carcinoma (ESCC) from three institutions were allocated into training (n = 80), internal validation (n = 35), and external validation (n = 71) cohorts. Clinical data were extracted from electronic medical records, and radiomic features were extracted from pretreatment contrast-enhanced CT images. Deep transfer learning features were extracted using residual network-50 (ResNet-50) and combined with radiomic features to construct deep transfer learning radiomics (DTLR) features. Single-modality models based on clinical characteristics, radiomic features, and DTLR features were developed, respectively. A multimodal fusion model was then constructed using decision-level fusion. Model performance was evaluated in terms of discrimination, calibration, and clinical utility, using metrics including the area under the receiver operating characteristic curve (AUC), accuracy (ACC), and sensitivity. Results The multimodal combined model achieved AUCs of 0.957 (95% confidence interval [CI]: 0.917–0.996), 0.922 (95% CI: 0.837–1.000), and 0.865 (95% CI: 0.783–0.948) in the training, internal validation and external validation cohorts, respectively. The combined model achieved higher AUC values than the clinical, radiomics, and DTLR models across all cohorts, although the incremental benefit over the DTLR model was not statistically significant in any cohort. Decision curve analysis demonstrated that the multimodal model provided greater clinical net benefit than each single-modality model. Conclusion The multimodal model integrating clinical, radiomics and DTLR features showed favorable performance for predicting EF after chemotherapy and/or radiotherapy in patients with ESCC. This model may facilitate the identification of high-risk patients and provide guidance for optimizing individualized treatment strategies.

Authors

Institutions

Publication Details

Journal
Frontiers in Oncology
Published
2026-11-16
DOI
https://doi.org/10.3389/fonc.2026.1910607
Primary Topic
Radiomics and Machine Learning in Medical Imaging
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Multimodal deep learning radiomics model for predicting esophageal fistula after chemotherapy and/or radiotherapy in esophageal squamous cell carcinoma: a multicenter study

Weixing Li, chenyu wang, Haohan Zhang, Chenyu Li et al.
2 links
Frontiers in Oncology
Radiomics and Machine Learning in Medical Imaging
article

Multimodal deep learning radiomics model for predicting esophageal fistula after chemotherapy and/or radiotherapy in esophageal squamous cell carcinoma: a multicenter study

Weixing Li, chenyu wang, Haohan Zhang, Chenyu Li, Junyan Yue, Cancan Guan, Xuezhi Li, Pei Li, Yichen Li, Yuqin Chen
article en

Abstract

Objectives This study aims to develop a multimodal model that integrates deep learning radiomics and clinical characteristics to predict the risk of esophageal fistula (EF) after chemotherapy and/or radiotherapy in patients. Methods In this multicentre retrospective study, 186 patients with esophageal squamous cell carcinoma (ESCC) from three institutions were allocated into training (n = 80), internal validation (n = 35), and external validation (n = 71) cohorts. Clinical data were extracted from electronic medical records, and radiomic features were extracted from pretreatment contrast-enhanced CT images. Deep transfer learning features were extracted using residual network-50 (ResNet-50) and combined with radiomic features to construct deep transfer learning radiomics (DTLR) features. Single-modality models based on clinical characteristics, radiomic features, and DTLR features were developed, respectively. A multimodal fusion model was then constructed using decision-level fusion. Model performance was evaluated in terms of discrimination, calibration, and clinical utility, using metrics including the area under the receiver operating characteristic curve (AUC), accuracy (ACC), and sensitivity. Results The multimodal combined model achieved AUCs of 0.957 (95% confidence interval [CI]: 0.917–0.996), 0.922 (95% CI: 0.837–1.000), and 0.865 (95% CI: 0.783–0.948) in the training, internal validation and external validation cohorts, respectively. The combined model achieved higher AUC values than the clinical, radiomics, and DTLR models across all cohorts, although the incremental benefit over the DTLR model was not statistically significant in any cohort. Decision curve analysis demonstrated that the multimodal model provided greater clinical net benefit than each single-modality model. Conclusion The multimodal model integrating clinical, radiomics and DTLR features showed favorable performance for predicting EF after chemotherapy and/or radiotherapy in patients with ESCC. This model may facilitate the identification of high-risk patients and provide guidance for optimizing individualized treatment strategies.

Frontiers in OncologyVol. 16
Wenzhou Medical University (CN), First Affiliated Hospital of Henan University of Science and Technology (CN), First Affiliated Hospital of Henan University of Traditional Chinese Medicine (CN), Xinxiang Central Hospital (CN), First Affiliated Hospital of Zhengzhou University (CN), Wenzhou Medical University Renji College (CN), Henan Medical University (CN)
Good health and well-being
Radiomics and Machine Learning in Medical Imaging
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon