Author(s):
1. Iram Shahzadi:
Federal Urdu University Islamabad,,Pakistan
Abstract:
In diabetics, diabetic retinopathy (DR) is the most common cause of vision loss.About 90% of the DR patients will be spared from visual loss if they are givenproper care, which dictates to design a DR model for identifying the differentseverity levels to provide better care. This research contributes to an improvedmethod for localizing, segmenting, and classifying various forms of DR lesions.The proposed designed consists of two modules, in module-I a novellocalization model is proposed with the combination of pre-trained resnet-18 andYOLOv3.that is constructed with the selection of optimal layers to localize theretinal lesions. The localized images are segmented using semantic segmentationmodel, is developed by the combination of 16 selected CNN layers. The modelis trained from the ground annotated masks and optimized learning parameters.The proposed model performance is evaluated on Grand-challenge IDRIDdataset and improved results are reported. Later, in module-II deep features arederived from the pre-trained Efficientnet-b0 model and optimized usingimproved parameters of the Genetic algorithm (GA) for the classification ofdifferent types of DR lesions such as normal-DR (0), mild (1), moderate (2), andsevere (3) on the Kaggle dataset. The proposed model achieved greater than98% classification accuracy that is better as compared to already publishedwork.
Page(s):
99-99
DOI:
DOI not available
Published:
Journal: 4th International Conference of Sciences “Revamped Scientific Outlook of 21st Century, 2025” , November 12,2025, Volume: 1, Issue: 1, Year: 2025
Keywords:
deep learning
,
Diabetic retinopathy
,
genetic algorithm
,
EfficientNetB0
,
ResNet18
,
semantic segmentation
,
YOLOv3
References:
References are not available for this document.
Citations
Citations are not available for this document.