MRI-based AI framework for predicting lymph node metastasis and prognosis in rectal cancer after neoadjuvant chemoradiotherapy: a multicenter study.
Authors
Affiliations (6)
Affiliations (6)
- Department of Radiology, The Sixth Affiliated Hospital, Sun Yat-Sen University, Guangzhou, China.
- Biomedical Innovation Center, The Sixth Affiliated Hospital, Sun Yat-sen University, Guangzhou, China.
- Department of Radiology, Guangdong Provincial People's Hospital, Guangzhou, China.
- Department of Radiology, The Third Affiliated Hospital, Sun Yat-sen University, Guangzhou, China.
- Department of Radiology, The Sixth Affiliated Hospital, Sun Yat-Sen University, Guangzhou, China. [email protected].
- Biomedical Innovation Center, The Sixth Affiliated Hospital, Sun Yat-sen University, Guangzhou, China. [email protected].
Abstract
Early prediction of lymph node metastasis (LNM) after neoadjuvant chemoradiotherapy (nCRT) is crucial for improving treatment planning and prognosis in rectal cancer (RC) patients. This study aims to develop an MRI-based AI framework integrating clinical, radiomics, and autoencoder features to predict LNM and prognosis post-nCRT. This study included 577 RC patients who underwent nCRT followed by surgical resection, with data collected from three centers. The dataset was divided into training set, internal validation set and external test set to develop a fully automated AI framework. The framework consists of a 3D U-Net lesion segmentation model, a feature extraction scheme integrating clinical, radiomics and autoencoder features, and an LNM classification model using machine learning. SHAP values elucidated the contributions of individual features and Cox regression analyses identified key prognostic features. Additionally, we compared the diagnostic capabilities of the proposed model with three radiologists in identifying LNM. The lesion segmentation model achieved a DICE coefficient of 0.81 on the external test set. The LNM classification models (Random Forest, LightGBM, CatBoost, XGBoost) showed AUROC values of 0.71, 0.71, 0.75, and 0.77 on the internal validation set, and 0.62, 0.67, 0.65, and 0.73 on the external test set, with XGBoost performing best. The model outperformed radiologists in accuracy, specificity, and sensitivity for LNM diagnosis. Higher values of T2WI_feat_16 and DWI_feat_164 were linked to poor prognosis. This AI framework provides a tool for predicting LNM and prognosis in RC patients after nCRT, enhancing clinical decision-making and prognostic assessment.