انسيه مرادي

Eye diseases, as abnormalities that affect the structures an‎d functions of vision, range from mild an‎d temporary disorders to chronic an‎d progressive conditions, which —if not diagnosed in time—may ultimately lead to permanent blindness. Diseases such as cataract, glaucoma, an‎d diabetic retinopathy can irreversibly threaten vision when their detection is delayed. Automated classification of these diseases based on fundus images, by enabling rapid diagnosis an‎d targeted treatment, can play a crucial role in transforming the early detection an‎d prevention of ocular disorders. In this thesis, with the aim of developing an optimized machine learning model, a hybrid framework named BayesSVM50 is introduced as an intelligent system for the automated classification of eye diseases from fundus images. In the first step, fundus images were preprocessed using contrast enhancement, normalization, an‎d size stan‎dardization techniques. Then, deep features were extracted using several pretrained convolutional neural networks, including VGG16, VGG19, ResNet50, EfficientNet, an‎d DenseNet. For dimensionality reduction an‎d selec‎tion of the most informative features, methods such as Principal Component Analysis (PCA), correlation analysis, variance thresholding, an‎d feature importance ranking were applied. The BayesSVM50 model, implemented as a support vector machine based on features extracted from the EfficientNet network, was tuned an‎d trained using Bayesian optimization. The results demonstrated that this model achieved an accuracy of 93.6 ± 1.05% after dimensionality reduction to 500 components using PCA, an‎d improved further to 95.3 ± 0.6% after optimization, outperforming classical approaches. The proposed approach, by integrating comprehensive feature engineering with Bayesian optimization, elevates the accuracy of eye disease classification from fundus images to a level comparable with state-of-the-art advanced models, indicating its potential for deployment in intelligent clinical screening systems.

ﺑﯿﻤﺎريﻫﺎي ﭼشمي , ﯾﺎدﮔﯿﺮي ﻣﺎﺷﯿﻦ , ﺗﺤﻠﯿﻞ ﻣﺆﻟﻔﻪﻫﺎي اصلي , ﺑﻬﯿﻨﻪﺳﺎزي ﺑﯿﺰي , تصاويرفوندوس

Eye Diseases , Machine Learning , Principal Component Analysis , Bayesian Optimization , Fundus Images

Ensieh Moradi

Prof. Jalil Rashidinia