Inception-Resnet V2 Based Eye Disease Classification Using Retinal Images

Abstract

Eye diseases are a diverse group of ocular conditions that can affect the visual health and overall wellbeing of individuals. The early detection and classification of eye diseases hold profound significance in the realm of healthcare and ophthalmology. Accurate classification of eye diseases enables healthcare providers to tailor treatment plans, monitor disease progression, and allocate resources efficiently. Furthermore, it facilitates the development of predictive models and the identification of risk factors, paving the way for personalized medicine and targeted preventive strategies. This paper presents an effective eye disease classification model based on Inception Resnet V2 model with fine-tuning mechanism. The Inception-ResnetV2, known for its exceptional feature extraction capabilities, is pretrained on a large-scale dataset and subsequently fine-tuned on a curated dataset of labeled eye disease images. The proposed system provided significant performance after fine-tuning. The model had 81.00% accuracy, which remarkably increased to 94.74% after fine-tuning. The enhanced accuracy of the model reflects improved precision and reliability, with precision increasing from 79.25% to 93.99%, reducing false positives, and recall rising from 82.49% to 94.24%, thereby lowering false negatives. Consequently, the F1-Score, which combines precision and recall, improved significantly from 81.49% to 94.11%. Finally, the model classifies the retinal images into four classes (Cataract, Glaucoma, Diabetic Retinopathy, Normal). This paper represents a significant step towards developing a reliable and efficient tool for the automated diagnosis and early detection of eye diseases, ultimately enhancing patient care in the field of ophthalmology.

Topics & Concepts