Development and validation of MRI-based radiomics models for predicting recurrence in patients with spine and pelvis chordomas.
Authors
Affiliations (8)
Affiliations (8)
- The Quzhou Affiliated Hospital of Wenzhou Medical University, Quzhou People's Hospital, Quzhou City, 324002, Zhejiang Province, China.
- Department of Orthopaedic Oncology, Spine Tumor Center, Changzheng Hospital of the Navy Medical University, No. 415 Fengyang Road, Shanghai, 200003, China.
- Department of Radiology, Changzheng Hospital of the Navy Medical University, No. 415 Fengyang Road, Shanghai, 200003, China.
- Department of Radiology, The First Naval Hospital of Southern Theater Command, Zhanjiang, 524005, Guangdong Province, China.
- Department of Orthopedics, Navy Medical Center, the Navy Medical University, No. 338 Huaihai West Road, Shanghai, 200052, China.
- Department of Radiology, Changzheng Hospital of the Navy Medical University, No. 415 Fengyang Road, Shanghai, 200003, China. [email protected].
- Department of Orthopaedic Oncology, Spine Tumor Center, Changzheng Hospital of the Navy Medical University, No. 415 Fengyang Road, Shanghai, 200003, China. [email protected].
- Department of Radiology, Changzheng Hospital of the Navy Medical University, No. 415 Fengyang Road, Shanghai, 200003, China. [email protected].
Abstract
Spine and pelvis chordoma (SPC) is a rare and aggressive mesenchymal tumor with poor long-term recurrence-free survival rates. Accurate preoperative prediction of recurrence remains a significant clinical challenge, and reliable prognostic methods are lacking. To develop and validate MRI-based radiomics models for predicting recurrence in SPC. A single-center retrospective study. A total of 135 SPC patients who underwent preoperative MRI, including T2-weighted imaging (T2WI) and contrast-enhanced T1-weighted (CET1) sequences. Recurrence status of SPC, evaluated using accuracy (ACC), sensitivity (SEN), specificity (SPE), area under the receiver operating characteristic curve (AUC), positive predictive value (PPV), and negative predictive value (NPV). A total of 2394 radiomic features were extracted from manually segmented MRI regions of interest. Features were selected using Spearman correlation analysis and LASSO regression with ten-fold cross-validation. Eleven machine learning algorithms were applied to construct radiomics models, and clinical predictors were identified via univariate and multivariate logistic regression. A combined model incorporating radiomics and clinical data was developed and visualized with a nomogram. Model performance was evaluated using ROC curves, decision curve analysis (DCA), calibration plots, and the DeLong test. Age, tumor size, and resection mode were identified as independent prognostic factors. Among radiomics-based models, the Random Forest (RF) model showed the best performance, with AUCs of 0.929 and 0.847 in the training and test cohorts, respectively, outperforming the clinical model with AUCs of 0.680 and 0.700. The combined model integrating radiomics and clinical features further improved performance, with AUCs of 0.945 and 0.897, and demonstrated the highest net clinical benefit according to DCA. The integrated model constructed by radiomics and clinical characteristics achieved robust preoperative recurrence prediction for SPC, offering personalized treatment and enhanced clinical decision-making.