Radiomics analysis of T1WI and T2WI magnetic resonance images to differentiate between IgG4-related ophthalmic disease and orbital MALT lymphoma

Abstract

Abstract Background Preoperative differentiation between IgG4-related orbital disease (IgG4-ROD) and orbital mucosa-associated lymphoid tissue (MALT) lymphoma has a significant impact on clinical decision-making. Our research aims to construct and evaluate a magnetic resonance imaging (MRI)-based radiomics model to assist clinicians to better identify IgG4-ROD and orbital MALT lymphoma and make better preoperative medical decisions. Methods MR images and clinical data from 20 IgG4-ROD patients and 30 orbital MALT lymphoma patients were classified into a training (21 MALT; 14 IgG4-ROD) or validation set (nine MALT; six IgG4-ROD). Radiomics features were collected from T1-weighted (T1WI) and T2-weighted images (T2WI). Student’s t-test, the least absolute shrinkage and selection operator (LASSO) and principal component analysis (PCA) were conducted to screen and select the radiomics features. Support vector machine (SVM) classifiers developed from the selected radiomic features for T1WI, T2WI and combined T1WI and T2WI were trained and tested on the training and validation set via five-fold cross-validation, respectively. Diagnostic performance of the classifiers were evaluated with area under the curve (AUC) readings of the receiver operating characteristic (ROC) curve, and readouts for precision, accuracy, recall and F1 score. Results Among 12 statistically significant features from T1WI, four were selected for SVM modelling after LASSO analysis. For T2WI, eight of 51 statistically significant features were analyzed by LASSO followed by PCA, with five features finally used for SVM. Combined analysis of T1WI and T2WI features selected two and four, respectively, for SVM. The AUC values for T1WI and T2WI classifiers separately were 0.722 ± 0.037 and 0.744 ± 0.027, respectively, while combined analysis of T1WI and T2WI classifiers further enhanced the classification performances with AUC values ranging from 0.727 to 0.821. Conclusion The radiomics model based on features from both T1WI and T2WI images is effective and promising for the differential diagnosis of IgG4-ROD and MALT lymphoma. More detailed radiomics features and advanced techniques should be considered to further explore the differences between these diseases.

Keywords

• Machine learning
• IgG4-ROD
• Orbital MALT lymphoma
• SVM