Résumé
To assess the performance of radiomics model based on cardiac magnetic resonance imaging (CMR) cine sequence for assessing the severity of mitral regurgitation. Methods: A total of 80 patients who underwent CMR and echocardiography examination were retrospectively enrolled, including 67 patients with no or slight mitral regurgitation and 13 patients with moderate or severe mitral regurgitation. The relative difference in average filtered gradient (RDAFG) of CMR cine sequence were generated, which were combined with minimum output sum of squared error tracker (MOSSE) to extract 25 radiomics features. After reducing feature dimensionality by principal component analysis (PCA) and oversampling the minority samples, the radiomics model was established using support vector machine (SVM). The performance of the model was assessed by receiver operating characteristic (ROC) curve. Results: There were significant differences (both P<0.01) of the 2-dimension radiomics features between the two groups. The best performance (area under the ROC curve) of the established radiomics model was 0.971, with sensitivity and specificity at 85.7% and 94.1%, respectively. Conclusion: The performance of the machine learning-based radiomics model derived from CMR cine sequence for assessing the severity of mitral regurgitation was excellent, which can facilitate the computer-aided diagnosis and treatment in the era of artificial intelligence.