Automated Disease Diagnostics using OCT and Deep Learning

Introduction: Retinal diseases such as Diabetic Macular Edema (DME) together with Choroidal Neovascularization (CNV) and Drusen qualify as the major contributors to blindness and vision deterioration across the globe which represents a major socio-medical issue [1]. Phase one detection remains essential to prevent serious health complications because proper interventions can follow at the right time. Optical Coherence Tomography (OCT) provides physicians with non-touch examination capabilities to view retinal diseases throughout their vertical dimensions. Because these images demonstrate complex characteristics their analysis needs advanced computational methods [2]. This study evaluates the compatibility of ResNet50 and VGG16 deep learning architectures for their application in OCT image classification tasks. The assessment depends on how well these networks identify complex retinal structures as well as detect diseases including DME alongside CNV and Drusen and determine accurate disease severity [3]. Methods: This research transforms OCT image data into four categories namely Drusen and CNV and DME together with Normal (projected healthy retina). Exponential normalization methods along with contrast enhancement techniques are applied to the acquired images. The feature extraction process utilizes VGG16 and ResNet50 models that undergo transfer learning for their optimization. The metrics accuracy and precision and recall and F1-score evaluate model performance according to [4]. Results: Tests show VGG16 outperforms ResNet50 in terms of obtaining higher precision rates alongside recall and accuracy measurements and F1-score. Due to its deeper architecture ResNet50 needed additional computational resources as well as longer processing times. The ophthalmological identification function in VGG16 proved more effective for recognizing diseases in retinal images [5]. Conclusions: The application of deep learning models VGG16 and ResNet50 enhances OCT image analysis which results in superior retinal disease classification frameworks according to research by [6]. Research results indicate VGG16 exceeds other models in OCT classification making it an appropriate selection for systems which detect retinal diseases automatically. Future development of disease detection models requires research into hybrid methods which use GLCM (Gray Level Co-occurrence Matrix) analysis methods to increase accuracy measurement abilities [7].