Machine Learning Based Non-Enhanced CT Radiomics for the Identification of Orbital Cavernous Venous Malformations: An Innovative Tool

Abstract

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.

Artificial Intelligence Applications in Chronic Ocular Diseases

“…Explainable artificial intelligence (XAI) is a potential solution to open the "black box" and gain the trust of patients. When detecting myopic macular lesions using OCT images, deep learning models can be trained using soft labels and output the probability of belonging to each lesion category rather than predicting a particular category, which has been shown to yield satisfying results (Du et al, 2022). Other visualization methods, such as the occlusion test (Zeiler and Fergus, 2014), saliency maps (Simonyan et al, 2013) and gradient-weighted class activation maps (Grad-CAMs) (Selvaraju et al, 2017) can also retrospectively analyze the prediction process of neural networks and highlight important regions relevant to decision making, thus improving interpretability.…”

End-to-End Deep-Learning-Based Diagnosis of Benign and Malignant Orbital Tumors on Computed Tomography Images

“…Guo et al [ 27 ] employed an MR-based radiomics signature to distinguish ocular adnexal lymphoma (OAL) and idiopathic orbital inflammation (IOI), which reached an AUC of 0.74 and 0.73 in primary and validation cohorts. Han et al [ 44 ] used an ML-based model to automatically identify the differences in orbital cavernous venous malformation from non-contrast-enhanced CT. Hou et al [ 28 ] reported a bag-of-features (BOF)-based radiomic analysis method, with a support vector machine as the classifier. Differentiation with augmentation achieved an AUC of 0.803 in the test group from contrast-enhanced MRI, but the same analysis results from pre-contrast MRI were significantly less reliable.…”

Orbital and eyelid diseases: The next breakthrough in artificial intelligence?

“…The developed neural network yielded an AUC of 0.979 in diagnosing patients with moderate-to-severe GO. Han et al (2022) automatically identified the differences in the orbital cavernous venous malformations (OCVM) from orbital CT images by training 13 ML models, including support vector machines (SVMs) and random forests. Nakagawa et al (2022) implemented a VGG-16 network to determine from CT images whether a nasal or sinus tumor invades the periorbital area.…”