Development of an Automated system for Eye Disease detection and classification using

Novel eyednet Model

Cataracts are often diagnosed by an eye care physician who uses a slit-lamp to evaluate structural

abnormalities in the eye lens (Wirbelauer, 2023). The doctor thereafter identifies and classifies

cataracts to ascertain their specific kind and degree of severity, enabling them to devise

appropriate treatment strategies (Thylefors et al., 2002). The classification and grading of

cataracts, particularly in their progressive age-related forms, have been extensively studied in

clinical and epidemiological research (Morales-Lopez, Cruz-Vega, and Rangel-Magdaleno,

2021). Cataracts are diagnosed by skilled eye care professionals utilizing slit-lamp

biomicroscopy and assessed by comparing the level of opacity in each slit-lamp picture with a

set of standard images (Yadav and Yadav, 2023). The standard pictures show a progressive level

of cataract severity, which is represented by ascending integer grades (Wong et al., 2013). The

ungraded slit-lamp picture is compared to standard images, and a decimal or integer grade is

assigned to indicate the degree of opacity (Huang et al., 2010; Chylack, 1993).

Several widely used grading protocols for assessing cataracts include Lens Opacities

Classification System (LOCS) III, Age-Related Eye Disease Study (AREDS) cataract grading

protocol, Wisconsin Grading System (WGS), Oxford Clinical Cataract Grading System

(OXCGS) (Srivastava et al., 2014), and World Health Organization (WHO) Grading System. The

LOCS-III and Wisconsin grading systems are widely used by eye care physicians (Shaheen and

Tariq, 2019). The LOCS III system evaluates four characteristics: nuclear color (NC), nuclear

opalescence (NO), posterior subcapsular cataract (P), and cortical cataract (C) (Chylack et al.,

1993). This diagnostic process necessitates the use of costly medical equipment. Additionally,

manual diagnostic procedures are both time-consuming and subjective, since they rely on the

clinician's own experiences (Srivastava et al., 2014). According to a study by Chylack et al.
(1993), when several eye care professionals are asked to assess the identical slit-lamp pictures

using the same grading system, they only get around 65% agreement across observers.

In addition, eye physicians are susceptible to unintentional and inaccurate assessment while

examining a large number of photographs (Guo, Yang, Peng, Li, and Liang, 2015). It is important

to note that accurately evaluating the hardness of cataracts is essential for minimizing surgical

problems (Xu et al., 2019). Choosing an inappropriate phacoemulsification energy level during

cataract surgery may disrupt the posterior capsule (Caixinha, Velte, Santos, and Santos, 2014).

Therefore, it is crucial to determine the ideal phacoemulsification energy level, which depends

on the density and grading of the cataract (Caixinha, Velte, Santos, and Santos, 2014). While it

now requires a skilled eye physician to assess the severity of a cataract in only a few minutes,

advancements in imaging technology have led researchers to suggest the creation of automated

systems that can identify specific features of the lens and provide scores for grading cataracts.

These locations are accurately associated with LOCS III clinical grades and save testing time.

(Gali, Sella, & Afshari, 2019). The rise of artificial intelligence (AI) in the field of

ophthalmology is driven by the expanding availability of large-scale clinical data (Goh et al.,

2020). Nevertheless, AI's proficiency and advancement in cataracts remain comparatively

restricted when compared to other ocular conditions such as diabetic retinopathy, age-related

macular degeneration, and glaucoma (Goh et al., 2020).

Prior studies have employed algorithms to automatically detect cataracts using different

techniques like slit lamp or color fundus. More recently, there has been a shift towards grading

cataracts using images captured with smartphones (Peterson, Ho, and Chong, 2020; Hu et al.,

2020; Tawfik, Birry, and Saad, 2018; Zhang et al., 2017). A gadget equipped with a sophisticated

camera and very vivid and high-definition displays has the potential to be used in the field of
ophthalmology and eye care (Lakshminarayanan, Zelek, and McBride, 2015). The global

presence of billions of smartphone users has created significant opportunities for transformative

advancements in healthcare applications. Smartphones are offering cost-effective and automated

options for costly medical tests, particularly in areas with limited access to medical personnel

and equipment. While the slit-lamp is often used for detecting and grading cataracts, employing

digital camera pictures for screening cataracts is preferred since it is simpler and easier to utilize.

The present research study aims to examine the process of automated cataract grading utilizing

photos captured by mobile phones, taking into account all the aforementioned factors.

Cataract is a prevalent ocular ailment worldwide and a prominent contributor to both blindness

and visual impairment. The conventional method for identifying cataracts entails an

ophthalmologist doing an eye examination using a slit-lamp microscope or ophthalmoscope.

During this examination, the ophthalmologist assesses the presence of cloudiness in the lens of

the eye, which is typically transparent. The global healthcare system faces challenges due to

limited resources and a shortage of qualified professionals. As a result, researchers are

investigating the potential of artificial intelligence (AI) solutions to support and augment the

expertise of healthcare professionals. Inspired by advancements in iris identification, this study

aims to introduces a new algorithm that combines many methods to identify cataracts using near-

infrared eye pictures.

The current work introduces a unique computer-aided automated cataract grading approach that

can identify different stages of cataracts, such as normal, early, pre-mature, and mature stages,

utilizing digital camera photos. The dataset from IIITD, consisting of 2380 pictures, is used

(Keshari et al., 2016). This dataset consists of two sessions: one before surgery with 145

individuals, and one after surgery with 99 subjects. The primary objective of the suggested
technique is to provide an automated cataract grading system that covers the whole process. The

suggested technique involves automated detection of the ocular area, which will then be retrieved

and used as inputs for a CNN model. The CNN approach classifies ungraded lenses into four

groups: normal, early, pre-mature, and mature.

Researchers have offered many methodologies for addressing the issue of illness identification.

One sort of study aimed to identify specific diseases in individuals, while another focused on

differentiating various phases of particular diseases depending on their severity. Recent research

has conducted the categorization of eye illnesses using both multi-label and multi-class

approaches. The study focuses on the multi-class categorization of four distinct classes. Initially,

researchers chose to use conventional machine learning methods such as Support Vector

Machines (SVM), K-Nearest-Neighbours (KNN), K-Singular Value Decomposition (K-SVD),

and Feature Vectors. Neural Networks 3 and Deep Learning were regarded as methods to expand

and enhance the outcomes of conventional machine learning techniques. Ensemble of various

deep learning models (Jiang et al., 2019) and Dense Correlation (DC) network (Li et al., 2020)

models have shown impressive performance in classification. The precise categorization of

pictures remains a difficult undertaking, particularly when it comes to classifying various eye

illnesses.

Hence, it is essential to find a solution for this sector that is both resilient and precise in its ability

to identify various eye ailments. This research proposes a novel method for classifying eye

diseases by using the Hybrid InceptionV3 and BiLSTM Network (EyeDNet). This involves

combining the benefits of Convolutional Neural Networks (CNN) with Bidirectional Long Short-

Term Memory (BiLSTM). In this scenario, CNN would serve as a feature extractor for the
fundus images, while BiLSTM would be used as a classifier. The main contributions of this work

is proposing as follows:

i. Proposing an automated approach using Hybrid InceptionV3 and BiLSTM (EyeDNet)

models for accurate and efficient classification of eye diseases.

ii. Combines the feature extraction capabilities of Convolutional Neural Networks (CNN)

with the classification power of Bidirectional Long Short-Term Memory (BiLSTM).

iii. Uses fundus images to validate the proposed method, demonstrating improved accuracy

over traditional machine-learning techniques.

iv. Compare the proposed study (performance with existing systems).

This proposed study is justified by the difficulties that currently exist in the diagnosis of

cataracts, such as the subjectivity and inconsistent nature of manual grading systems, the need

for costly equipment and skilled doctors, and the restricted availability of healthcare resources in

some areas. Through the creation of a computer-aided grading system, this research proposes to

provide solution to a significant gap in the state of healthcare today.

References

Wirbelauer, C. (2023). Cataract forms and grading. In Cataract and Lens Surgery (pp. 21-30).

Cham: Springer International Publishing.

Thylefors, B., Chylack Jr, L. T., Konyama, K., Sasaki, K., Sperduto, R., Taylor, H. R., & West4,

S. (2002). A simplified cataract grading system The WHO Cataract Grading Group. Ophthalmic

epidemiology, 9(2), 83-95.

classification systems using computational intelligence: A survey. Archives of Computational

Methods in Engineering, 28, 1761-1774.

Yadav, S., & Yadav, J. K. P. S. (2023). Automatic Cataract Severity Detection and Grading Using

Deep Learning. Journal of Sensors, 2023.

Wong, W. L., Li, X., Li, J., Cheng, C. Y., Lamoureux, E. L., Wang, J. J., ... & Wong, T. Y. (2013).

Cataract conversion assessment using lens opacity classification system III and Wisconsin

cataract grading system. Investigative ophthalmology & visual science, 54(1), 280-287.

Huang, W., Chan, K. L., Li, H., Lim, J. H., Liu, J., & Wong, T. Y. (2010). A computer assisted

method for nuclear cataract grading from slit-lamp images using ranking. IEEE Transactions on

Medical Imaging, 30(1), 94-107.

Chylack, L. T., Wolfe, J. K., Singer, D. M., Leske, M. C., Bullimore, M. A., Bailey, I. L., ... &

Wu, S. Y. (1993). The lens opacities classification system III. Archives of ophthalmology, 111(6),

831-836.

Srivastava, R., Gao, X., Yin, F., Wong, D. W., Liu, J., Cheung, C. Y., & Wong, T. Y. (2014).

Automatic nuclear cataract grading using image gradients. Journal of Medical Imaging, 1(1),

014502-014502.

Guo, L., Yang, J. J., Peng, L., Li, J., & Liang, Q. (2015). A computer-aided healthcare system for

cataract classification and grading based on fundus image analysis. Computers in Industry, 69,

72-80.
Xu, C., Zhu, X., He, W., Lu, Y., He, X., Shang, Z., ... & Li, X. (2019). Fully deep learning for

slit-lamp photo based nuclear cataract grading. In Medical Image Computing and Computer

Assisted Intervention–MICCAI 2019: 22nd International Conference, Shenzhen, China, October

13–17, 2019, Proceedings, Part IV 22 (pp. 513-521). Springer International Publishing.

Caixinha, M., Velte, E., Santos, M., & Santos, J. B. (2014, September). New approach for

objective cataract classification based on ultrasound techniques using multiclass SVM classifiers.

In 2014 IEEE International Ultrasonics Symposium (pp. 2402-2405). IEEE.

Gali, H. E., Sella, R., & Afshari, N. A. (2019). Cataract grading systems: a review of past and

present. Current opinion in ophthalmology, 30(1), 13-18.

Goh, J. H. L., Lim, Z. W., Fang, X., Anees, A., Nusinovici, S., Rim, T. H., ... & Tham, Y. C.

(2020). Artificial intelligence for cataract detection and management. Asia-Pacific journal of

ophthalmology, 9(2), 88-95.

Peterson, D., Ho, P., & Chong, J. (2020). Detecting Cataract Using Smartphone. Investigative

Ophthalmology & Visual Science, 61(7), 474-474.

Hu, S., Wu, H., Luan, X., Wang, Z., Adu, M., Wang, X., ... & He, W. (2020). Portable handheld

slit-lamp based on a smartphone camera for cataract screening. Journal of Ophthalmology, 2020.

Tawfik, H. R., Birry, R. A., & Saad, A. A. (2018). Early recognition and grading of cataract using

a combined log Gabor/discrete wavelet transform with ANN and SVM. International Journal of

Computer and Information Engineering, 12(12), 1038-1043.

detection and grading using deep convolutional neural network. In 2017 IEEE 14th international

conference on networking, sensing and control (ICNSC) (pp. 60-65). IEEE.

Lakshminarayanan, V., Zelek, J., & McBride, A. (2015). " Smartphone Science" in Eye Care and

Medicine. Optics and Photonics News, 26(1), 44-51.

Keshari, R., Ghosh, S., Agarwal, A., Singh, R., & Vatsa, M. (2016, September). Mobile

periocular matching with pre-post cataract surgery. In 2016 IEEE International Conference on

Image Processing (ICIP) (pp. 3116-3120). IEEE.