MRI-based traditional radiomics and computer-vision nomogram for predicting lymphovascular space invasion in
Ling Long1, Jianqing Sun2, Liling Jiang1
1Department of Radiology, Chongqing University Cancer Hospital & Chongqing Cancer Institute & Chongqing Cancer Hospital, 400030 Chongqing, PR China.
Purpose:
To determine the capabilities of MRI-based traditional radiomics and computer-vision (CV) nomogram for predicting lymphovascular space invasion (LVSI) in patients with endometrial carcinoma (EC).
Materials And Methods:
A total of 184 women (mean age, 52.9±9.0 [SD] years; range, 28-82 years) with EC were retrospectively included. Traditional radiomics features and CV features were extracted from preoperative T2-weighted and dynamic contrast-enhanced MR images. Two models (Model 1, the radiomics model; Model 2, adding CV radiomics signature into the Model 1) were built. The performance of the models was evaluated by the area under the curve (AUC) of the receiver operator characteristic (ROC) in the training and test cohorts. A nomogram based on clinicopathological metrics and radiomics signatures was developed. The predictive performance of the nomogram was assessed by AUC of the ROC in the training and test cohorts.
Results:
For predicting LVSI, the AUC values of Model 1 in the training and test cohorts were 0.79 (95% confidence interval [CI]: 0.702-0.889; accuracy: 65.9%; sensitivity: 88.8%; specificity: 57.8%) and 0.75 (95% CI: 0.585-0.914; accuracy: 69.5%; sensitivity: 85.7%; specificity: 62.5%), respectively. The AUC values of Model 2 in the training and test cohorts were 0.93 (95% CI: 0.875-0.991; accuracy: 94.9%; sensitivity: 91.6%; specificity: 96.0%) and 0.81 (95% CI: 0.666-0.962; accuracy: 71.7%; sensitivity: 92.8%; specificity: 62.5%), respectively. The discriminative ability of Model 2 was significantly improved compared to Model 1 (Net Reclassification Improvement [NRI]=0.21; P=0.04). Based on histologic grade, FIGO stage, Rad-score and CV-score, AUC values of the nomogram to predict LVSI in the training and test cohorts were 0.98 (95% CI: 0.955-1; accuracy: 91.6%; sensitivity: 91.6%; specificity: 96.0%) and 0.92 (95% CI: 0.823-1; accuracy: 91.3%; sensitivity: 78.5%; specificity: 96.8%), respectively.
Conclusions:
MRI-based traditional radiomics and computer-vision nomogram are useful for preoperative risk stratification in patients with EC and may facilitate better clinical decision-making.
