Among orbital lymphoproliferative disorders, about 55% of diagnosed cancerous tumors are orbital lymphomas, and nearly 50% of benign cases are immunoglobulin G4-related ophthalmic disease (IgG4-ROD). However, due to nonspecific characteristics, the differentiation of the two diseases is challenging. In this study, conventional magnetic resonance imaging-based radiomics approaches were explored for clinical recognition of orbital lymphomas and IgG4-ROD. We investigated the value of radiomics features of axial T1- (T1WI-) and T2-weighted (T2WI), contrast-enhanced T1WI in axial (CE-T1WI) and coronal (CE-T1WI-cor) planes, and 78 patients (orbital lymphoma, 36; IgG4-ROD, 42) were retrospectively reviewed. The mass lesions were manually annotated and represented with 99 features. The performance of elastic net-based radiomics models using single or multiple modalities with or without feature selection was compared. The demographic features showed orbital lymphoma patients were significantly older than IgG4-ROD patients ( p < 0.01 ), and most of the patients were male (72% in the orbital lymphoma group vs. 23% in the IgG4-ROD group; p = 0.03 ). The MR imaging findings revealed orbital lymphomas were mostly unilateral (81%, p = 0.02 ) and wrapped eyeballs or optic nerves frequently (78%, p = 0.02 ). In addition, orbital lymphomas showed isointense in T1WI (100%, p < 0.01 ), and IgG4-ROD was isointense (60%, p < 0.01 ) or hyperintense (40%, p < 0.01 ) in T1WI with well-defined shape (64%, p < 0.01 ). The experimental comparison indicated that using CE-T1WI radiomics features achieved superior results, and the features in combination with CE-T1WI-cor features and the feature preselection method could further improve the classification performance. In conclusion, this study comparatively analyzed orbital lymphoma and IgG4-ROD from demographic features, MR imaging findings, and radiomics features. It might deepen our understanding and benefit disease management.

Purpose: To evaluate the capability of non-enhanced computed tomography (CT) images for distinguishing between orbital cavernous venous malformations (OCVM) and non-OCVM, and to identify the optimal model from radiomics-based machine learning (ML) algorithms. Methods: A total of 215 cases of OCVM and 120 cases of non-OCVM were retrospectively analyzed in this study. A stratified random sample of 268 patients (80%) was used as the training set (172 OCVM and 96 non-OCVM); the remaining data were used as the testing set. Six feature selection techniques and thirteen ML models were evaluated to construct an optimal classification model. Results: There were statistically significant differences between the OCVM and non-OCVM groups in the density and tumor location (P < 0.05), whereas other indicators were comparable (age, gender, sharp, P > 0.05). Linear regression (area under the curve [AUC] ¼ 0.9351; accuracy ¼ 0.8657) and Stochastic Gradient Descent (AUC ¼ 0.9448; accuracy ¼ 0.8806) classifiers, both of which coupled with the f test and L1-based feature selection method, achieved optimal performance. The support vector machine (AUC ¼ 0.9186; accuracy ¼ 0.8806), Random Forest (AUC ¼ 0.9288; accuracy ¼ 0.8507) and eXtreme Gradient Boosting (AUC ¼ 0.9147; accuracy ¼ 0.8507) classifier combined with f test method showed excellent average performance among our study, respectively. Conclusions:The effect of non-enhanced CT images in OCVM not only can help ophthalmologist to find and locate lesion, but also bring great help for the qualitative diagnosis value using radiomicbased ML algorithms.

Background and purposeInverted papilloma (IP) and nasal polyp (NP), as two benign lesions, are difficult to distinguish on MRI imaging and clinically, especially in predicting whether the olfactory nerve is damaged, which is an important aspect of treatment and prognosis. We plan to establish a new biomarker to distinguish IP and NP that may invade the olfactory nerve, and to analyze its diagnostic efficacy.Materials and methodsA total of 74 cases of IP and 55 cases of NP were collected. A total of 80% of 129 patients were used as the training set (59 IP and 44 NP); the remaining were used as the testing set. As a multimodal study (two MRI sequences and clinical indicators), preoperative MR images including T2-weighted magnetic resonance imaging (T2-WI) and contrast-enhanced T1-weighted magnetic resonance imaging (CE-T1WI) were collected. Radiomic features were extracted from MR images. Then, the least absolute shrinkage and selection operator (LASSO) regression method was used to decrease the high degree of redundancy and irrelevance. Subsequently, the radiomics model is constructed by the rad scoring formula. The area under the curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of the model have been calculated. Finally, the decision curve analysis (DCA) is used to evaluate the clinical practicability of the model.ResultsThere were significant differences in age, nasal bleeding, and hyposmia between the two lesions (p &lt; 0.05). In total, 1,906 radiomic features were extracted from T2-WI and CE-T1WI images. After feature selection, using 12 key features to bulid model. AUC, sensitivity, specificity, and accuracy on the testing cohort of the optimal model were, respectively, 0.9121, 0.828, 0.9091, and 0.899. AUC on the testing cohort of the optimal model was 0.9121; in addition, sensitivity, specificity, and accuracy were, respectively, 0.828, 0.9091, and 0.899.ConclusionA new biomarker combining multimodal MRI radiomics and clinical indicators can effectively distinguish between IP and NP that may invade the olfactory nerve, which can provide a valuable decision basis for individualized treatment.