DOIONLINE

DOIONLINE NO - IJASEAT-IRAJ-DOIONLINE-19970

Publish In
International Journal of Advances in Science, Engineering and Technology(IJASEAT)-IJASEAT
Journal Home
Volume Issue
Issue
Volume-11,Issue-2  ( Apr, 2023 )
Paper Title
Efficient Net-B5 Based Transfer Learning for Multi-Label Classification of Diabetic Retinopathy Using CLAHE Preprocessing
Author Name
Mehul Jain, Mustafa Noman Rashid, Priyanka Aggarwal, Kapil Sharma
Affilition
Pages
121-127
Abstract
Diabetes mellitus, a chronic disorder that is rapidly becoming more prevalent worldwide. It is defined by hyperglycemia, which restricts the body's utilization of glucose. Diabetic retinopathy (DR), which comes in four severity levels—mild, moderate, severe, and proliferative—is a frequent side effect of diabetes. Deep learning is used in this study to categorize the intensity of diabetic retinopathy using neural networks and transfer learning. The APTOS 2019 Blindness Detection dataset is used in this work. Several pre-processing techniques are used, including resizing, grayscale conversion, Gaussian blur, thresholding, contour detection, CLAHE, and median filtering, to improve the picture quality and make it easier to identify and segment retinal lesions.Convolutional neural networks (CNNs) are used with the EfficientNet-B5 architecture for feature extraction and classification. Transfer learning is used for training the model, and accuracy isbeing used as the performance parameter for evaluation. With an accuracy of 98.77%, on the APTOS-2019 dataset demonstrate the efficacy of the recommended technique. These encouraging findings demonstrate the possibility of neural network-based methods for categorizing an imbalanced dataset for diabetic retinopathy. The suggested framework can offer highly accurate models that are appropriate for clinical data from real-world patients and diagnostic applications, helping to identify and treat diabetic retinopathy. Keywords - Diabetic Retinopathy Classification, Aptos, Clahe, Efficientnet-B5, Convolutional Neural Network, Transfer Learning
  View Paper