Multi-Stage Classification of Retinal OCT Using Multi-Scale Ensemble Deep Architecture

Accurate noninvasive diagnosis of retinal disorders is required for appropriate treatment or precision medicine. This work proposes a multi-stage classification network built on multi-scale (pyramidal) feature ensemble architecture image using optical coherence tomography (OCT) images. First, scale-adaptive neural developed to produce inputs extraction and learning. The larger input sizes yield more global information, while the smaller focus local details. Then, feature-rich pyramidal designed extract features as DenseNet backbone. advantage hierarchical structure that it allows system multi-scale, information-rich accurate disorders. Evaluation two public OCT datasets containing normal abnormal retinas (e.g., diabetic macular edema (DME), choroidal neovascularization (CNV), age-related degeneration (AMD), Drusen) comparison against recent networks demonstrates advantages proposed architecture’s ability with average accuracy 97.78%, 96.83%, 94.26% first (binary) stage, second (three-class) all-at-once (four-class) classification, respectively, cross-validation experiments dataset. In dataset, our showed an overall accuracy, sensitivity, specificity 99.69%, 99.71%, 99.87%, respectively. Overall, tangible enhanced learning might be used in various medical tasks where scale-invariant are crucial precise diagnosis.