Optical Coherence Tomography Image Classification using Light-weight Hybrid Transformers

Diabetic retinopathy is a diabetes complication that affects the retina of the eye. Conditions like Hyperglycemia and Diabetes can damage the blood vessels in the retina, causing vision problems or even blindness. Early on, the condition often has no symptoms, but it can be detected through regular eye exams. The paper focuses on identifying the various cases of this impairment like Diabetic Macular Edema (DME) and Agerelated Macular Degeneration (AMD) implications like Choroidal Neovascularization (CNV) and Drusen, present in Optical Coherence Tomography images using very lightweight, data-efficient, CNN-based transformer, namely MobileVit. The classification results were obtained using the MobileVit-XXS, the lightest variant of the MobileVit. A balanced, publicly accessible dataset was used to train the model, which was then fine-tuned for optimum performance. This work proposes a CAD methodology using a lightweight CNN-based Transformer network. The accuracy generated by the model is 98.86% and the F1-score is 93.50%. A simple application is developed to test the deployability of the model.