A Hybrid Lightweight ResNet50-1D-CNN Transfer Learning Model for Accurate Eye Disease Classification
7th Doctoral Symposium on Intelligence Enabled Research, DoSIER 2025, Kolkata, India, 21 - 22 November 2025, vol.1959 LNNS, pp.424-436, (Full Text)
- Publication Type: Conference Paper / Full Text
- Volume: 1959 LNNS
- Doi Number: 10.1007/978-3-032-25446-7_35
- City: Kolkata
- Country: India
- Page Numbers: pp.424-436
- Keywords: CNN, Eye Disease classification, ResNet50, Transfer learning
- Ataturk University Affiliated: Yes
Abstract
In medical applications where computer-aided diagnosis is beneficial, classification of eye disease detection is crucial. Inaccurate diagnosis of eye diseases will reduce the patient’s chances of survival and hinder their ability to respond to the administered treatment. This work presents an automated approach for identifying various eye disorders using an ensemble of deep transfer learning to address these drawbacks. The target task was used to fine-tune five pre-trained convolutional neural network (CNN) architectures: ResNet-50, InceptionV3, and EfficientNet (B1-B3). The suggested model used the Adaptive Gradient (Adagrad) optimizer to modify the network weights to minimize the loss function. A total of 4217 images were subjected to this algorithm; 1038 of these images showed cataracts, 1098 showed diabetic retinopathy, 1007 showed glaucoma, and 1074 showed normal. Accuracy metrics and the area under the curve (AUC) served as performance indicators. The success rate of the suggested transfer learning techniques is comparable to that of previous research; the best classification result, achieved with ResNet50 using Adadelta, is 98.1%. As a result, the transfer learning model has promise for the medical field and can assist physicians in making prompt and precise judgments.