Abstract

     Glaucoma is one of the most common chronic diseases that may lead to irreversible vision loss. The number of patients with permanent vision loss due to glaucoma is expected to increase at an alarming rate in the near future. A considerable amount of research is being conducted on computer-aided diagnosis for glaucoma. Segmentation of the optic cup (OC) and optic disc (OD) is usually performed to distinguish glaucomatous and non-glaucomatous cases in retinal fundus images. However, the OC boundaries are quite non-distinctive; consequently, the accurate segmentation of the OC is substantially challenging, and the OD segmentation performance also needs to be improved. To overcome this problem, we propose two networks, separable linked segmentation network (SLS-Net) and separable linked segmentation residual network (SLSR-Net), for accurate pixel-wise segmentation of the OC and OD. In SLS-Net and SLSR-Net, a large final feature map can be maintained in our networks, which enhances the OC and OD segmentation performance by minimizing the spatial information loss. SLSR-Net employs external residual connections for feature empowerment. Both proposed networks comprise a separable convolutional link to enhance computational efficiency and reduce the cost of network. Even with a few trainable parameters, the proposed architecture is capable of providing high segmentation accuracy.

Introduction

     Glaucoma has become one of the major causes of vision loss, and in this disease, the optic nerve head (ONH) is damaged (Tham et al., 2014). Glaucoma causes gradual vision loss, and the patient has no abrupt considerable symptoms; hence, its early detection and screening are crucial. Many advanced imaging methods are employed by experts for retinal disease diagnosis and assessment. Fundus imaging is widely used in glaucoma detection tasks because it is fast, affordable, and non-invasive (Edupuganti et al., 2018). Color fundus imaging best serves the glaucoma detection in both advanced glaucoma or early glaucoma detection cases (Ahn et al., 2018). Fundus imaging also enables researchers and experts for computational analysis like cup-to-disc ratio (CDR) computation which significantly helps in glaucoma detection (Orlando et al., 2020).

     Several methods have been used for the assessment of glaucoma; however, owing to numerous clinical and resource problems, they could not fill the gap of its early diagnosis (Baum et al., 1995). Compared to other methods, the ONH assessment is more commonly used. Automated ONH assessment methods are gaining popularity over manual methods these days because the

Conclusion

     In this research, we proposed deep learning-based novel models, SLS-Net and SLSR-Net, to segment OD and OC for glaucoma screening. SLSR-Net is the final proposed model that maintains a large final feature map size throughout the network to avoid spatial information loss of even minor features. Memory requirements are one of the major limitations of computer-aided diagnosis. An SCL unit in our model minimizes this problem and significantly increases the computational efficiency of the network. External residual connections settle the feature degradation problem by empowering the features. Training and testing of the network is carried out without any preprocessing or postprocessing overhead. We extensively evaluated the proposed model on four publicly available datasets and achieved state-of-the-art performances compared with existing methods. There is a trade-off between segmentation performance and computational efficiency. Therefore, methods that achieve good segmentation performance usually use many parameters, which makes the network computationally expensive. In our network, good results are achieved by using only 4,666,950 trainable parameters, which confirms the outstanding computational efficiency of the network compared to the state-of-the-art methods.