Value of MR-based radiomics in differentiating uveal melanoma from other intraocular masses in adults. Issue 131 (October 2020)

• Record Type: Journal Article
• Authors: Su, Yaping; Xu, Xiaolin; Zuo, Panli; Xia, Yuwei; Qu, Xiaoxia; Chen, Qinghua; Guo, Jian; Wei, Wenbin; Xian, Junfang
• Is Part Of: European journal of radiology. Issue 131(2020)
• Publication Date: 2020-10
• Subjects: Uveal melanoma -- Magnetic resonance imaging (MRI) -- Machine learning -- Radiomics -- Differential diagnosis; Medical radiology -- Periodicals; Radiology -- Periodicals; Radiologie médicale -- Périodiques; Medical radiology; Periodicals; 616.075705
• DOI: 10.1016/j.ejrad.2020.109268
• Languages: English
• ISSNs: 0720-048X

Abstract

Highlights: The treatment of UM is quite different from that of the other intraocular masses and accurate diagnosis is vital for optimizing treatment. All machine learning classifiers performed well, especially in the combined T2WI and CET1WI model, where accuracy and AUC ranged from 76 % to 86 % and from 0.870 to 0.877, the sensitivity and specificity ranged from 66.7 %–93.9 % and from 70.6 %–94.1 % in the test set, and the MLP classifier obtained the best discriminant effect for the diagnosis of UM. In the combined model, the performance of ML classifiers was better than the performance of visual assessment in the training set and in all patients (all p＜0.05). MR-based radiomics features can be a promising and second-read tool in the differential diagnosis of UM in adults. Abstract: Purpose: To assess the performance of machine learning (ML)-based magnetic resonance imaging (MRI) radiomics analysis for discriminating between uveal melanoma (UM) and other intraocular masses. Methods: This retrospective study analyzed 245 patients with intraocular masses (165 UMs and 80 other intraocular masses). Radiomics features were extracted from T1WI, T2WI, and contrast enhanced T1-weighted images (CET1WI), respectively. The intraclass correlation coefficient (ICC) was calculated to quantify the reproducibility of features. The training and test sets consisted of 195 and 50 cases. Least absolute shrinkage and selection operator (LASSO) regression method was employed for feature selection. The ML classifiers were logistic regression (LR), multilayer perceptron (MLP), and support vector machine (SVM). The performance of classifiers was evaluated by ROC analysis, and was compared to the performance of visual assessment by DeLong test. Results: The optimal radiomics feature set was 10, 15, 15, and 24 for T1W, T2W, CET1W, and joint T2W and CET1W images, respectively. The accuracy of T1WI, T2WI, CET1WI, and the joint T2WI and CET1WI models ranged from 72.0 %–78.0 %, from 79.6 %–81.6 %, from 74.0 %–82.0 %, and from 76.0 %–86.0 % in the test set. In the test set, the AUC for T1WI, T2WI, CET1WI, joint T2WI, and CET1WI models ranged from 0.775 to 0.829, 0.816 to 0.826, 0.836 to 0.861, and 0.870 to 0.877, respectively. In the combined model, the performance of ML classifiers was better than the performance of visual assessment in the training set and in all patients (p＜0.05). Conclusions: Radiomics analysis represents a promising tool in separating UM from other intraocular masses.