Fusion of MRI-radiomics and clinical features to improve pathological T-stage prediction of rectal cancer: a multicenter study.
Authors
Affiliations (6)
Affiliations (6)
- School of Health Science and Engineering, University of Shanghai for Science and Technology, Shanghai, 200093, China.
- Department of Radiology, Fudan University Shanghai Cancer Center, Shanghai, 200032, China. [email protected].
- Department of Oncology, Shanghai Medical College, Fudan University, Shanghai, 200032, China. [email protected].
- Department of Radiology, Fudan University Shanghai Cancer Center, Shanghai, 200032, China. [email protected].
- Department of Oncology, Shanghai Medical College, Fudan University, Shanghai, 200032, China. [email protected].
- Medical Imaging Center, The Affiliated Taian City Central Hospital of Qingdao University, Taian, Shandong, 271000, China. [email protected].
Abstract
Developing a rational model for precise preoperative staging of rectal cancer using magnetic resonance imaging is crucial for improving the therapeutic effect and prognosis of patients. To investigate the feasibility and efficacy of a T2-weighted imaging (T2WI) based radiomics model combined with clinical features for preoperative T1-2/T3 staging prediction in rectal cancer. A total of 968 rectal adenocarcinoma patients (T1-2 stage, n = 562; T3 stage, n = 406) were retrospectively enrolled in this study. We randomly divided it into a training cohort (n = 774) and an internal validation cohort (n = 194). Additional 82 patients from another center used as an independent external validation cohort. For each patient, 1210 radiomics features were extracted from T2WI images. Least absolute shrinkage and selection operator were applied for feature selection. Support vector machine classifier was used to develop the radiomics model. Six clinical features were collected and selected to build the clinical model using the recursive feature elimination algorithm. A weighted fusion method was applied to combine the radiomics with clinical model to develop the fusion model. The performances of the above three models were evaluated using ROC, calibration, and decision curves. The radiomics and clinical model yielded AUC of 0.858 (95% CI, 0.801-0.904) and 0.780 (95% CI, 0.715-0.836) in internal validation cohort, while the fusion model generated higher performance with AUC of 0.875 (95% CI, 0.820-0.918). The external validation cohort further validated the effectiveness of the fusion model with AUC of 0.828 (95% CI, 0.729-0.903). The calibration and decision curves confirmed that the fusion model was well calibrated and had clinical utility. Combined T2WI radiomics features and clinical features modeling can effectively predict T-stage (T1-2/T3) of rectal cancer and provide valuable reference for clinical decision making.