|
Choroidal neovascularization (CNV) is a significant complication associated with age-related macular degeneration (AMD), particularly its "wet" or exudative form. In the context of AMD, CNV is the leading cause of severe vision loss and legal blindness. Detecting CNV can prove challenging, particularly in its early stages when symptoms may be subtle, or the abnormal blood vessel growth is minimal. Modern diagnostic tools, such as Optical Coherence Tomography (OCT) and Fluorescein Angiography (FA), have greatly improved our ability to diagnose CNV through comprehensive eye examinations. In this paper, we discuss the classification of normal and CNV affected retina optical coherence tomography (OCT) images, aiming to improve diagnostic reliability and assist medical professionals, working in the field of Ophthalmology. CNV is characterized by the growth of new choroidal vessels into the sub-retinal space through breaks in the Bruch's membrane. Our classification model combines image pre-processing, feature extraction, and image classification techniques. It leverages both traditional machine learning methods like Support Vector Machines (SVM), Random Forest, and AdaBoosting, as well as advanced deep learning architectures like Convolutional Neural Networks (CNN) and the highly regarded VGG16. After an extensive evaluation of model performance, the VGG16-SVM hybrid model emerged as the top performer, achieving an outstanding 99% accuracy rate and an equally impressive 99% precision rate. This result highlights the model's robustness and effectiveness in accurately categorizing images. The superiority of the VGG16-SVM hybrid model can be attributed to VGG16's proficiency in hierarchical feature learning. By combining the strengths of deep learning and traditional machine learning, the model harnesses deep networks' exceptional feature extraction capabilities and the robust classification power of Support Vector Machines (SVM). This synergy results in a more balanced and precise classification performance.
|
