A cataract, or the clouding (opacity) of the lens, is one of the most common eye diseases prevalent in 51% of all eye patients [1]. According to the World Health Organization (WHO), cataracts will impact approximately 93 million people in the year 2021. It causes mild to severe vision loss in over 80% of patients, eventually causing permanent blindness in severer cases [2]. Amongst other reasons, age and diabetes are considered the primary factors for the disease. It is also often caused due to injury to the eye as well, such as blunt force trauma to the eye [3]. The initial symptoms are not as intense, leading to delayed disease identification. Blurry vision and headaches are some more prominent symptoms [4]. Early detection and proper identification can often lead to the prevention of permanent blindness.
Based on a study by the WHO, cataracts can be classified by the location of the opacity; there are three kinds of cataracts of various severity, known as the sclerotic nuclear cataract, posterior subcapsular cataract, and the cortical spoking cataract. The differences in the cataract are apparent and noticeable in the images shown in Fig. 1. The nuclear cataract is the existence of opacity in the central optical region known as the nucleus, visible in the Fig. 1a. The posterior subcapsular is the presence of the opacity in the posterior region, as shown in Fig. 1b. Finally, the presence of a spoke-like structure throughout the lens is known as the cortical spoking cataract, shown in Fig. 1c [1].
The proposed method implements the classification process in multiple stages. The images undergo various image processing techniques to detect the presence of cataracts. The deep learning model identifies the type of cataract in the diseased eye. The main incentive behind this project is to address the everyday issue people face due to the lack of knowledge of this disease, as it can be deceptive. The most alluring part of our method is that it is simple, portable, and stress-free.
In recent years, researchers have had notable progress in developing algorithms for automatically detecting cataracts [5]. Cataracts are classified in various methods. The detection is done using image processing and machine learning techniques[6]. Recently, researchers are also making significant progress in classifying the type of cataract using deep learning techniques. Despite making progress, little research has been done on grading the type of cataracts [references].
In 2020, [7] classified and graded cataracts in stages. They used the Indian Institute of Technology, Delhi (IITD) dataset, which contained both pre-and post-operational eye images. Among several approaches, the most significant was using a Support Vector Machine. they achieved an accuracy of 87%. They also performed landmark detection to detect the eyes, trained pre-trained networks, and used data augmentation to improve accuracy. In [8], CNN based deep learning network has been implemented using DenseNet and InceptionNet to classify the cataract types. The Age-Related Eye Disease Study (AREDS) dataset was used, which is a labeled dataset with 576 images.
In this paper, we propose a cataract detection method that combines image processing and machine learning techniques to identify and classify cataracts. The cataract is detected using the image processing technique, following which the grading (type) is done with the machine learning model, which correctly identifies the type of cataract (nuclear, posterior, or spoking cataract).