International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 11 Issue: 09 | Sep 2024
p-ISSN: 2395-0072
www.irjet.net
AGCRNL: Automatic Glaucoma Classification Using Residual Network and LSTM Jaswanth N 1 1SRM University ---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - In this work, the diagnosis of glaucoma in the
the drainage system by being extremely close to it. It requires immediate attention and can be severely painful.
early stage using deep learning models has been presented. Glaucoma is a group of diseases which is irreversible and can cause blindness. Manual diagnosis depends on human skills and it is difficult to diagnose at an early stage. Glaucoma classification using deep learning is a challenging task. The concept of feature extraction using deep learning models has been utilized in the current study. An ensemble residual network along with squeeze and excitation block and with BiLSTM (Binary Long Short-Term Memory) has been implemented for sequential feature extraction for faster and more accurate classification of the disease. A residual network is used for global feature extraction. The accuracy rate of the validation set was 92.3% for our model. The findings suggest that a cost-effective screening tool for early and cost-effective identification of glaucoma could be developed utilizing deep learning algorithms.
Manual analysis of glaucoma can be time-consuming and cumbersome. The accuracy of parameter measurement for detection varies according to the doctors. In recent years, various automated techniques and diversity indices have been proposed for the automatic detection of the disease using retinal fundus images [5]. The Residual Network 50 (ResNet50) model has proven effective for the extraction of features and classifying the fundus image [6]. The accuracy of the model was found to be 93%. The residual network has been used in numerous research for feature extraction including detection of colorectal cancer and insect pest recognition [7, 8]. Taking motivation from the research of Ovreiu et. al. (2020) [9], Residual with squeeze and excitation block has been combined in this research to obtain channelwise recalibrated features. It reduces and discards features based on channel re-calibration and helps to train the model in an efficient way.
Key Words: Bi-LSTM, Glaucoma Classification, ResNet50, SE block.
In this article, we propose a novel architecture for fully automated binary classification of glaucoma where the Fundus database has been utilized for training and testing purposes. Different databases (e.g., ORIGA, ACRIMA and Fundus) can also be used for glaucoma classification. In the proposed framework, a residual network has been combined with a squeeze and excitation block followed by a BiLSTM block [10]. In addition, different available methods for glaucoma classification have been compared with the proposed model.
1.INTRODUCTION Glaucoma is the second progressive worldwide leading cause of eye diseases. The early-stage detection of glaucoma is a challenging task. Currently, around 79.6 million individuals are affected by glaucoma [1] and it has been estimated that by 2040 the number of affected individuals will be around 111.8 million [2]. The eye is one of the most complex, sensitive and delicate organs which is responsible for passing visual information to the brain cells in the human body. The root cause of glaucoma is the elevated pressure in the eye which can be associated with Intra-Ocular Pressure (IOP) [3]. Improper drainage of the liquid aqueous humor leads to an increase in pressure in the eye subsequently affecting the retina and optic nerve and human vision. In general, the IOP value should be less than 21mmHg to fight against glaucoma. Manual diagnosis is based on structural changes of the retinal nerve fibre layer and optic nerve head [4]. Primarily, glaucoma can broadly be classified into two categories: 1) Open-angle glaucoma and 2) Angleclosure glaucoma. The effect of Openangle glaucoma disease is not realized until it starts to impair human vision [2]. This type of glaucoma disease is painless and is caused when the drainage canal is not able to drain out the excess aqueous humor. People with high blood pressure and diabetes are more susceptible towards this type of glaucoma. In the Angle-closure glaucoma disease, the iris tends to block
© 2024, IRJET
|
Impact Factor value: 8.315
Fig -1: Diagram of normal eye and eye with glaucoma [11] MOTIVATION Glaucoma disease is the second leading cause of blindness globally [12]. The critical work of identifying and categorizing glaucoma is challenging which motivated us to
|
ISO 9001:2008 Certified Journal
|
Page 435