Neoadjuvant Chemoradiotherapy on Postoperative Complications of Rectal Cancer: A Retrospective Study Integrating MRI Radiomics and Deep Learning.
Authors
Affiliations (6)
Affiliations (6)
- Department of Anesthesiology and Perioperative Medicine, The First Affiliated Hospital with Nanjing Medical University, Nanjing, 210029, China.
- Department of Radiation Oncology, The Third Affiliated Hospital of Soochow University, Changzhou, 213003, China.
- Department of Radiation Oncology, The First Affiliated Hospital with Nanjing Medical University, Nanjing, 210029, China.
- Department of Radiation Oncology, The First Affiliated Hospital with Nanjing Medical University, Nanjing, 210029, China. [email protected].
- Department of Anesthesiology and Perioperative Medicine, The First Affiliated Hospital with Nanjing Medical University, Nanjing, 210029, China. [email protected].
- Department of Radiation Oncology, The First Affiliated Hospital with Nanjing Medical University, Nanjing, 210029, China. [email protected].
Abstract
Neoadjuvant chemoradiotherapy (nCRT) is standard for locally advanced rectal cancer but increases postoperative complication (POC) risks due to tissue fibrosis and immunosuppression. Existing models based on clinical parameters overlook nCRT-induced tissue changes. Radiomics and deep learning technologies have been widely applied in tumor-related predictive models in recent years, but their potential in predicting postoperative complications has not yet been fully explored. This study aimed to investigate the impact of neoadjuvant chemoradiotherapy on postoperative complications in rectal cancer patients and develop a deep learning-based predictive model integrating MRI radiomics and clinical features for early risk stratification. A retrospective, dual-center study included 695 patients, comprising 272 nCRT patients and 423 non-nCRT patients. Propensity score matching (PSM) balanced baseline covariates. Radiomics and deep learning features were extracted from preoperative MRI of the nCRT group, respectively. Six machine learning algorithms were evaluated for predictive performance. nCRT was identified as a risk factor for POC (OR 3.78, P < 0.001). The radiomics-deep learning combined model outperformed unimodal approaches, with an AUC of 0.82 in external validation. In conclusion, nCRT is a risk factor of POC in LARC patients and the multimodal deep learning model shows superior predictive accuracy, providing a clinically actionable tool for personalized perioperative management.