CT imaging-based radiomics and deep learning models for predicting chemotherapy response in advanced pancreatic cancer
4 affiliations
- 1Department of Radiology, Affiliated Hospital of Nanjing University of Chinese Medicine, Jiangsu Province Hospital of Chinese Medicine
- 2Department of Radiology, Xiangyang First People’s Hospital
- 3Department of Radiology, The Second Affiliated Hospital of Nanjing Medical University
- 4Department of Mathematics, Nanjing University
Abstract
To investigate the value of radiomics and deep learning features derived from pre-treatment CT imaging in predicting the efficacy of chemotherapy in patients with advanced pancreatic cancer. The retrospective study included 207 patients with advanced pancreatic cancer from two medical centers, divided into a training cohort (n = 139) and an external validation cohort (n = 68). Radiomics features were extracted from portal-venous-phase CT images and selected via Least Absolute Shrinkage and Selection Operator (LASSO) regression. A variational autoencoder (VAE) extracted deep learning features from three-dimensional tumor volumes. The clinical, radiomics, and combined models were constructed and compared using the DeLong test. Twenty-three radiomics features were selected for model construction. In the training cohort, the radiomics model (AUC = 0.953) and the combined model (AUC = 0.965) outperformed the clinical model (AUC = 0.765, P < 0.001). In the external validation cohort, the radiomics model achieved an AUC of 0.982, with a precision of 0.85, a recall of 0.83, and an F1-score of 0.84 for identifying responders. Multivariate analysis identified cT and cN stages as independent predictors of treatment response. VAE-based deep features maintained prediction accuracy of 0.75–0.82 across different classifier types. CT-based radiomics models can predict chemotherapy response in advanced pancreatic cancer, outperforming clinical models. VAE-derived deep learning features also show promising predictive performance.
Related
- Fusion of Deep Learning Radiomics and Clinical Features for Individualized Prediction of Neoadjuvant Chemotherapy Response in Osteosarcoma
- Magnetic resonance imaging radiomics for predicting pathologic complete response to neoadjuvant chemotherapy in breast cancer
- Mammography based delta radiomics with machine learning predicts pathologic complete response after neoadjuvant chemotherapy in breast cancer
- A multimodal deep learning radiomics nomogram integrating clinicopathological features for predicting pathological complete response after neoadjuvant therapy in triple-negative breast cancer: a multicenter retrospective study
- Incremental Value and Cross-Center Validation of Pretreatment Tri-Modal Ultrasound Deep Learning for Predicting Breast Pathologic Complete Response to Neoadjuvant Chemotherapy: A Multicenter Study
BibTeX
@article{zhu2026imagingbased,
title = {{CT imaging-based radiomics and deep learning models for predicting chemotherapy response in advanced pancreatic cancer}},
author = {Yong Zhu and Hao Zhou and Peng An and Yingfan Mao and Ziwei Nie and Jianhua Wang and Zhongqiu Wang and Wenjing Cui},
journal = {Scientific Reports},
year = {2026},
month = jul,
doi = {10.1038/s41598-026-62816-5},
url = {https://doi.org/10.1038/s41598-026-62816-5},
note = {CC BY},
}
RIS
TY - JOUR
TI - CT imaging-based radiomics and deep learning models for predicting chemotherapy response in advanced pancreatic cancer
AU - Yong Zhu
AU - Hao Zhou
AU - Peng An
AU - Yingfan Mao
AU - Ziwei Nie
AU - Jianhua Wang
AU - Zhongqiu Wang
AU - Wenjing Cui
PY - 2026
DA - 2026/07/01
JO - Scientific Reports
DO - 10.1038/s41598-026-62816-5
UR - https://doi.org/10.1038/s41598-026-62816-5
AB - To investigate the value of radiomics and deep learning features derived from pre-treatment CT imaging in predicting the efficacy of chemotherapy in patients with advanced pancreatic cancer. The retrospective study included 207 patients with advanced pancreatic cancer from two medical centers, divided into a training cohort (n = 139) and an external validation cohort (n = 68). Radiomics features were extracted from portal-venous-phase CT images and selected via Least Absolute Shrinkage and Selection Operator (LASSO) regression. A variational autoencoder (VAE) extracted deep learning features from three-dimensional tumor volumes. The clinical, radiomics, and combined models were constructed and compared using the DeLong test. Twenty-three radiomics features were selected for model construction. In the training cohort, the radiomics model (AUC = 0.953) and the combined model (AUC = 0.965) outperformed the clinical model (AUC = 0.765, P < 0.001). In the external validation cohort, the radiomics model achieved an AUC of 0.982, with a precision of 0.85, a recall of 0.83, and an F1-score of 0.84 for identifying responders. Multivariate analysis identified cT and cN stages as independent predictors of treatment response. VAE-based deep features maintained prediction accuracy of 0.75–0.82 across different classifier types. CT-based radiomics models can predict chemotherapy response in advanced pancreatic cancer, outperforming clinical models. VAE-derived deep learning features also show promising predictive performance.
SN - 2045-2322
ER -