Automatic optic disc segmentation for edema classification and severity grading

Optic disc edema (ODE) is a notable ocular symptom often seen in neuro-ophthalmic illnesses affecting the optic disc (OD). The etiologies underlying ODE are diverse, including a range of symptoms and consequences. The early detection of ODE can mitigate the risk of vision loss and the development of severe ocular disorders. The texture seen in the edematous OD exhibits a notable dissimilarity when compared to the non-edematous OD in retinal images. Consequently, methods usually effective for non-edematous cases may exhibit limited efficacy when applied to edematous situations. In this study, we provide a comprehensive approach for the automated classification of OD fundus photographs into edematous and non-edematous categories. Our proposed method is designed to handle fundus image datasets that combine both edematous and non-edematous ODs. The algorithm includes the processes of localizing, segmenting, and classifying optic discs that exhibit edematous and non-edematous characteristics. The studies proposed the factorized gradient vector flow (FGVF) method to do the segmentation of the edematous OD and non-edematous OD. The classification of the OD type is performed by using a linear support vector machine (SVM) algorithm, which is based on 27 distinct features that have been extracted from the vessels, GLCM, color, and intensity line profile. The approach method was evaluated using a total of 295 images, including 146 edematous cases and 149 non-edematous cases, sourced from three datasets. The segmentation process demonstrates an average accuracy of 88.41%, recall of 89.35%, and F1-Score of 86.53%. The average classification accuracy is 99.40%, achieving a superior performance of 3.43% compared to the state-of-the-art method. The classification of ODE severity is important to provide appropriate therapies and minimize the risk of exacerbating outcomes. In the second study, a set of 15 appearance-based features were used, which were obtained from the localized ODE area, to assess the severity stage. The features are extracted from GLCM, color, intensity line profile, and vessels. The linear support vector machine (SVM) is used for the classification of a total of 146 ODE funds photographs that have been chosen from three datasets. The proposed method in this study shows favorable results, with accuracy, specificity, and sensitivity achieving values of 78.77%, 82.5%, and 74.24%, respectively. The proposed technique substantially improves over the state-of-the-art approach, achieving performance gains of 17.13%, 16.62%, and 18.5% in the corresponding evaluation metrics.