Literature Review
Early detection of glaucoma is essential on multiple levels. First, since the condition is linked to several serious side effects, including potential blindness, detecting it in early stages may be the only possible solution to preventing some of the common negative implications. Researchers point out that it is primarily dangerous because of the inability to detect it early since symptoms are often noticed too late, and the risk of blindness is high because of this particular reason (Carrillo et al., 2019). Thus, individuals with glaucoma may be unaware of their condition until the optic nerve damage is too severe and the abnormalities are irreversible. However, it is essential to mention that while there are ways to detect glaucoma as soon as the initial symptoms have appeared, there are several challenges that have to be minimized. Since the condition itself is complex and not fully understood in terms of detecting and addressing symptoms, an early diagnosis is critical (Baudouin et al., 2021). Thus, the research on the field of glaucoma is mostly centered around ways to adequately confront it when it is present, yet the main work revolves around detection in the early stages.
Needless to say, a condition that can cause blindness if no intervention is performed is better managed when detected when the symptoms are not critical. Lusthaus and Goldberg (2019) point out that it is irreversible, yet it can be addressed through pressure management. Certain measures can be implemented to avoid the most major risk, which is blindness. The intervention depends on the stage of glaucoma, its type, and the risks that it imposes on the patient. For example, researchers point out that the newest strategy in confronting the condition is the use of stem cells and nanoparticles, which have been used alongside standard surgical procedures (Kwon et al., 2020). These may become the new replacement of more conventional intervention due to the positive outcomes that have been observed so far. Invasive medicine, however, is not the only solution to disease management since it primarily addresses more severe cases that have not been diagnosed early on and impose a more significant threat to the well-being of the individual.
Certain drugs can also help in managing the condition depending on circumstances and the type of glaucoma. However, certain studies highlight that the deposition of certain substances at the ocular site can be damaging and cause cellular damage, which is ultimately harmful to one’s vision (Yadav, Rajpurohit and Sharma, 2019). Thus, drugs such as pills and drops are not always the safest way of confronting glaucoma. In order for the limitations to be addressed, the medical community is frequently showcasing new technological advancements in both diagnosing and addressing glaucoma. An example is the newly implemented iStent injection implantation method, which has been shown to be less damaging than other intervention methods such as medications and standard surgical procedures used on people with glaucoma (Clement et al., 2019). Still, the technique is relatively new, and more research has to be done on the subject for the medical community to be confident in using this method as a primary intervention measure. Since the procedure is new and the long-term effects have not been studied due to the lack of data, it is early to say that this is the revolutionary way of confronting one of the primary causes of blindness internationally.
The most effective way of addressing glaucoma, regardless of its type, is early detection. This is why extensive research has been done in regards to investigative work and effective methods for diagnosing glaucoma that will be cost-effective, comfortable for patients to engage in, and accurate in regards to the rate of correct diagnosis. This is especially important at the beginning of condition development when the signs are not as visible and highlighted.
There are different types of glaucoma, yet certain forms are more common, which means that contributing to research when it comes to addressing these particular conditions is essential. Researchers have found that the most common form, based on epidemiological studies, is open-angle glaucoma (Tanito et al., 2018). Open-angle glaucoma is illustrated through increased eye pressure caused by the ineffective work of the drainage canals. As the name implies, it has a wider angle between the cornea of the eye and the iris, yet it is dangerous since symptoms are often not noticed by individuals, causing them to address the issue in later stages when the damage is often irreversible. Another less common form is Angle-Closure glaucoma, alongside Normal-Tension Glaucoma, which does not result in increased pressure and may be harder to diagnose due to the lack of information on what causes it. A form of glaucoma that affects young children due to the incorrect development of the eye is congenital glaucoma. Some of the types that can also be classified as either open-angle or closed-angle glaucoma include secondary glaucoma, pigmentary glaucoma, pseudoexfoliative glaucoma, traumatic glaucoma, neovascular glaucoma, Irido Corneal Endothelial Syndrome (ICE), and uveitic glaucoma (Schuster et al., 2020). The classification depends on the progression of the disease, how it manifests, and its origins. Both automatic and non-machine detection can determine the form of glaucoma, yet specific difficulties suggest using the automated method as more accurate, hence, more beneficial.
Glaucoma Detection
The limitations are present in both automated and non-machine methods of detection, which are two of the ways used for identifying the condition. The non-machine way is still a prominent assessment tool that implies that the physician determines the diagnosis without implementing the use of machine learning, which is becoming more widely applied in glaucoma detection. However, researchers point out that the screening, which implies the expertise of the healthcare provider, is highly dependent on who is examining the eyesight. This includes such screening techniques as direct and indirect ophthalmoscopy, slit-lamp biomicroscopy, optical coherence tomography (OCT), and retinal photography performed by ophthalmologists (International Council of Ophthalmology, 2017). The measures have various objectives that can help healthcare providers detect whether the patient has glaucoma or other health issues. For example, optical coherence tomography is often used to determine the thickness of the retinal layer, which is lower for individuals with glaucoma compared to those with no such diagnosis (Maetschke et al., 2019). Thus, the procedure is often applied to either diagnose or monitor the condition in regards to worsening or improvement in certain symptoms. Needless to say, the methodologies mentioned prior are not ineffective in terms of detecting and preventing the adverse side effects of untreated glaucoma. However, as the field of medicine is constantly improving in regards to new technology, painless procedures, and patient-centred care, the disadvantages correlating with some of the screening models suggest that the current system can be effectively addressed through newer techniques.
The use of images is widely applied in detecting signs of glaucoma since it is an accurate assessment of the condition of the patient. However, the two major fields where pictures are used are automated and non-automated detection techniques. The non-automated, as mentioned prior, rely on the expertise of the doctor who is to determine the signs of the condition based on the images. Thus, the medical provider achieves the dilation of the pupils by using eye drops before performing further procedures (Hark et al., 2018). Pupil dilation is an effective way for doctors to have better access to the possible signs of glaucoma, which is why it is widely used in manual diagnosis procedures. However, the difference between the two approaches is the involvement of the medical professional when it comes to detecting the abnormalities. This, inevitably, correlates with certain negative factors. The main one is the subjectivity of glaucoma detection, which is highly reliant on the expertise of the physician. Furthermore, researchers specifically refer to clinicians interpreting disc photographs, which is a procedure that has been shown to vary depending on the doctor’s experience and knowledge in the field (Myers et al., 2018). Needless to say, the fact that non-machine methods of diagnosing glaucoma can be subjective is not the only downside. Another one is the fact that the process may be uncomfortable for the patient, which can be one of the causes why glaucoma is still a significant cause of blindness worldwide. Determining the condition and monitoring the progress of treatment can become a major inconvenience. For example, Kumar, Francesco and Sharma (2019) refer to gonioscopy as challenging for ophthalmologists to perform yet highly effective glaucoma detection measure. It is often used alongside a slit-lamp, a technique mentioned prior, to determine the angle between the cornea and iris. The procedure itself is challenging to perform, and when it comes to estimating the angle, physicians are often not skilled enough to make accurate assumptions.
It is also essential to mention that there are several imaging modalities that are practiced in medicine and can help experts determine signs of optic nerve damage. This includes confocal scanning laser ophthalmoscopy, optical coherence tomography, Heidelberg retina tomography, and fundus imaging (Septiarini et al., 2018). The imaging modality determines the structure of the retina, which is vital for detecting glaucoma and its type. However, as mentioned before, non-machine detection depends on the physician who examines the images, which can be deeply subjective. The ineffective diagnosis methods performed by ophthalmologists do not necessarily imply a lack of knowledge on the subject or an inadequate approach to evaluating the current condition of the patient. Instead, the optic disc structure and size differs based on age, ethnicity, and other factors that are difficult to take into consideration for experts when investigating the eyes (Rodriguez-Una and Azuara-Blanco, 2019). However, the machine-learning technique does not have the same human disruptors, such as the inability to detect small changes or remember specific characteristics which apply to a particular population.
In terms of the need for an expert to determine the condition of the retina and the characteristics, which will later affect the patient’s well-being in case of an intervention, another downside is the cost. This is another inconvenience that may deter specific individuals from visiting a specialized facility to conclude the state of the eyes. Thus, researchers mention that eliminating the experts as a grading system can significantly reduce the cost of procedures related to glaucoma detection (Tan et al., 2020). The pricing of current methodologies of diagnosis procedures and further addressing of glaucoma either through surgery or medication is one of the factors that contribute to such a high rate of blindness caused by optic nerve damage (Sahlsten et al., 2019). Research studies point out that the US alone spends almost $3 billion dollars a year for addressing glaucoma-related complications (Gan et al., 2020). The limitation is even more critical in terms of global spending. Moreover, certain countries that have fewer financial and medical opportunities to provide care for individuals with glaucoma are not capable of investing a significant amount of money into the research, diagnosis, and treatment of such patients.
Since the limitations of the non-medical modalities reduce the rate of people seeking medical help for a condition that can lead to blindness, other techniques are being developed that can improve the situation and, perhaps, address the healthcare challenge. Placing all the responsibility on the expertise of a physician can be unreliable in terms of the subjective opinion of the expert, the knowledge in the field, and the methodologies of assessing the risk or presence of glaucoma. As a result, an automated technique can become a major improvement that is not only safer but also has the potential of diagnosing more cases, which ultimately leads to an overall reduction in the worsening of the condition in the population worldwide. Needless to say, addressing the cost of needing the opinion of the doctor is critical, especially in developing countries. Such techniques can be performed without the presence of a doctor, which means that people in more remote locations or without the financial or physical abilities to visit medical facilities can access help. Furthermore, automated glaucoma detection does not imply the lack of an expert either performing the investigative procedures or using the results of supplementary information. Researchers illustrate that machine-learning glaucoma detection can also become a tool used by experts in deciding whether the automatic system has effectively selected the diagnosis or the pattern in morphological changes (Civit-Masot et al., 2020). Thus, ophthalmologists can see it as a supplementary assessment technique, yet it does not imply that the automated method cannot be disregarded by physicians when performing the treatment. This can be an accurate detection method that doctors can either agree or disagree with, yet its necessity is significant due to the challenges that medical professionals meet when it comes to non-machine glaucoma diagnosis.
Automated Glaucoma Detection
Automated glaucoma detection is a methodology that relies on certain technological tools that can assess the current condition of the eyes and determine whether the patient is in the early stages of glaucoma, has a more advanced condition, or does not have any signs of damage to the optic nerve. Since the system relies on automatic identification, the need for an expert is limited based on the technique that is being used. Researchers point out that there are three types of diagnostics tools for automated glaucoma examination. Those are based on using Fundus, Optical Coherence Tomography, and Confocal Scan Laser Tomography (Khalil et al., 2017). Moreover, the automated technique is divided into two categories based on the purpose of the examination and the characteristics that are being investigated. Thus, both functional and structural features can be examined during the diagnosis practice. While the functional features are centred around the performance of the eyes, the structural ones examine specific parts, their size, and certain abnormalities or deformations. In one case, the aim is to identify problems in the vision process, pressure, and other functional processes. In terms of structural eye examination based on automatic models, the cup and the disc are being observed. The segmentation of optic disc and cup, as illustrated in the relevant literature on the topic of ophthalmology, is one of the essential steps in glaucoma screening (Jiang, Tan and Peng, 2019). Furthermore, the automated approach can provide a more accurate result in terms of the ratio, which can significantly improve the state in which the patient decided to address the condition. Since glaucoma can be effectively confronted in the early stages, a methodology that can give a comprehensive and accurate medical answer early on before major symptoms occur is critical to be implemented.
Researchers apply different techniques to the segmentation of optic disc and optic cup. Yu et al. (2019) have applied the U-Net method to get image segmentation. This method was used to minimize the issues related to the use of large fundus images during machine learning analysis. Thus, the researchers have found that applying the U-Net method to the optic disc and cup segmentation is a productive way of maximining the potential of deep learning while disregarding the pieces of information redundant to the study. Thus, the technique involves a selective examination of information, which is always beneficial due to the lack of redundant data and a contribution to the overall efficacy of the diagnosis process.
The automatic approach is highly reliant on fundus photography since images of the rear of the eye can show the state of the retina and optic disc. According to researchers, this is typically done with the purpose of examining the cup-to-disc ratio (CDR) (Al-Bander et al., 2018). Fundus images are critically important in the field of automatic glaucoma detection. It has been proven to be more effective than other techniques that are widely used in ophthalmology. For example, Kausu et al. (2018) have conducted research that compared the results of the use of fundus images compared to raised intraocular pressure assessment in terms of the examination of the optic nerve, which is the essence of glaucoma. Based on the study’s findings, photographs of the back of the eye are more accurate in assessing the stage or presence of glaucoma than the non-automatic measures.
The automatic approach in glaucoma detection is also widely researched due to the increase of interest in telemedicine. Since the recent pandemic has shown that remote medical help can be effective and comfortable for patients depending on the conditions, more researchers are looking for ways to implement glaucoma detection and care in remote healthcare provision. Thus, a study illustrating the use of telemedicine for glaucoma diagnosis has been referring to the efficacy of the method (Hark et al., 2017). During the study, the participants provided two fundus pictures and one anterior segment photograph that were captured with a hand-held camera. Moreover, information such as the family history of eye conditions and lifestyle characteristics were introduced in a system before being examined by a trained expert remotely. While machine learning was not involved in the procedure, automatic measures illustrate that glaucoma detection can become a much less complex procedure for the patient. Moreover, the research refers to the necessity of advancement in the field due to the potential benefit for monitoring the progression of the diseases. Thus, patients can undergo the easy steps to provide the physicians with information or their symptoms, which will then be assessed remotely.
It is important to note that while imagery has been proven to be useful for experts in being more effective with making the right diagnosis, the interpretation can differ based on the expert. Thus, as mentioned prior, may be affected by several factors, such as the expertise of the physician on the level of experience in dealing with individuals with different types of glaucoma. As a result, researchers highlight that interpretation of disc photography is not always accurate since human error or bias may interfere with the results (Myers, Fudemberg and Lee, 2018). Due to the severity of the condition, such a variable tool may not be adequate in addressing the critical outcomes of undetected glaucoma that has not been appropriately addressed in the early stages when medication can contribute to the betterment of the situation. The issue can be minimized by implementing the use of machine learning techniques.
Machine Learning
Machine learning relies on the algorithms that are designed to improve a system through automatic data examination. Based on this technique, multiple images of different types of glaucoma on different stages are introduced for the system to detect the pattern. Moreover, images of healthy eyes with no diagnosed glaucoma can be used as a comparison factor, which can allow the differences to be highlighted. Researchers mention this to be one of the most accurate ways of proficiently analyzing a vast amount of information and detecting even small changes that doctors can easily miss (Serte and Serener, 2019). Based on this technique, the images of patients that will be introduced will be automatically assessed in terms of abnormalities, which are illustrated by the contrast with the pictures of healthy eyes and similarities with the images of individuals with glaucoma. Thus, the categories “normal” and “abnormal” are compiled using existing databases of images with both glaucoma and healthy eyes (Thakur and Juneja, 2020). The machine learning process implied the analysis of the existing information that will later serve as an algorithm used to assess new information based on the detected patterns and morphological characteristics.
The images that are most widely used in machine learning glaucoma detection include fundus images and optical coherence tomography. According to researchers, the models used for algorithms establishment differ, yet it does not imply they cannot be used together for more accurate assessment. For example, Wu et al. (2021) suggest that using optical coherence tomography (OCT), even the slightest changes in the structure of the optic nerve can be detected, which is the main feature of diagnosing glaucoma at an early stage. Thus, the benefit of using OCT for machine learning is the ability to assess glaucoma before major symptoms appear. Since, according to researchers, nerve degeneration is irreversible, OCT may be the solution to identifying the condition before any significant damage occurs (Saba et al., 2018). Other studies also support the use of OCT and artificial intelligence in glaucoma diagnosis. Recent research has shown that this approach is highly accurate in pinpointing structural changes and allowing medical providers to intervene in the early stages of glaucoma (Prabhakar, Singh and Yadav, 2021). However, fundus images are also primary tools in machine-learning techniques.
Fundus images allow for a view of the optic disc, hence, the optic cup. One of the main concerns in assessing the presence, lack, or severity of glaucoma is the optic disc-to-cup ratio. This, for example, was done by Fatima Bokhari et al. (2017) using fundus images, which are effective in determining whether the ratio is within the normal range or slightly abnormal. Glaucoma is illustrated through an increase in the ratio, which is detectable through fundus images. Thus, the machine learning technique can detect a higher vertical ratio, which is a common glaucoma symptom, using fundus images (Wang et al., 2019). The method has been found to be effective in determining the stage of the condition before the damage starts becoming dangerous.
Moreover, fundus images show other signs of glaucoma that are effectively interpreted both through machine learning and by experts in the field. Sarhan, Rokne and Alhajj (2019) refer to this method as being the most commonly used to detect the degeneration of retinal ganglion cells. This is another major factor that highlights the presence of structural changes due to the presence of glaucoma. Thus, fundus images show a wide array of signs that depict the condition and its stage. Moreover, to maximize the detection rate and minimize the risk of ineffective diagnosis, machine learning can be used alongside fundus images. In this case, fundus images act as information that is examined through deep learning for further determination of healthy versus eyes with glaucoma without the need for an expert.
The machine learning technique has been applied by multiple researchers, who have implemented applied various methodologies of assessment, data collection, and examination of glaucoma. For example, a study was performed while relying on a mixed approach including both fundus images and three-dimensional optical coherence tomography, which are two of the most reliable ways of detecting optical deteriorations (An et al., 2019). The researchers found that using the two approaches in one singular framework leads to effective glaucoma detection. The technique has helped develop a system in which the condition can be detected at early stages. This is possible due to the promising results, which have shown the sensitivity of the system in terms of differentiating the images between the control subjects and individuals with glaucoma. The quantified images that were used (Fundus and OCT) were examined through machine learning, and the conclusion refers to the accuracy of the technique.
Another research has examined glaucoma detection with the use of fundus images through wavelet-based denoising and machine learning. Thus, the researchers have constructed an algorithm through which the images are not only assessed in terms of the presence or the lack of glaucoma symptoms but also the characteristics of the pictures that can compromise the results, namely contrast, mean homogeneity, variance, and other possible disruptors (Khan et al., 2021). The results have been examined, and the authors of the study have noticed that the accuracy of the classification method has increased by more than 90%. This retinal image synthesis method has been shown to be influential in determining whether the image can be adequately examined through machine learning of certain characteristics that may disqualify it.
Thus, the approaches can be classified into two major categories, which are deep-learning and non-deep learning methods. While both are automatic, the deep-learning technique relies on more complex algorithms, while non-deep learning is more simplistic and linear. However, both methods require data entry, which can be problematic. This occurs either due to the lack of an extensive database of pictures with glaucoma versus without the condition, which creates difficulties in designing a system that can effectively address the extensive amount of information entered since deviations may disrupt the results. The two machine-learning limitations have been effectively addressed by researchers.
Researchers mention an effective way of reducing the amount of data while still maintaining an extensive database that can then be used for machine-learning algorithms. Thus, a systematic review conducted by Barros et al. (2020), relevant literature suggests a reduction in dimensionality and feature extraction for images of glaucoma. Hence, there are fewer data to process while the number of cases stays the same. In this case, an extensive database can be introduced, yet the processing time and amount become more manageable, which is beneficial for the machine learning system. The algorithm looks for certain features while avoiding the examination of the entire picture. Moreover, the whole process relies on less information without a significant reduction in content necessary for the investigation, which then leads to effective diagnosing.
The method was used in several pieces of research, and the authors have found it to be proficient in glaucoma detection. For example, Mitra et al. (2018) referred to the importance of having an area of interest, which can be included in the algorithm. The intent is to give the system fewer data to work on while keeping the necessary amount of useful information, such as the optic cum-to-disc ratio that can be examined without the analysis of additional features. Thus, the technique can be designed to detect changes in certain structures while avoiding the examination of the whole picture.
In terms of a lack of the necessary database of images required for machine learning, the solution is using altered images. Researchers have found a solution, which consists of using synthetic fundus images for machine learning data (Castro, Valenti and Tegolo, 2020). Based on the proposed method, instead of selecting images with glaucoma in different patients, the findings suggest the use of synthetic photographs as being compelling data. Thus, since specific attributes of the retinal blood vessels, such as the widths, corners, and lengths, can be helpful in determining glaucoma, slightly manipulating them in images can be interpreted as new information. Based on this framework, the photos can be altered based on the regular progression of the condition itself, which leads to morphological changes. Such changes do not have to be documented, yet a manual alteration is enough for the machine learning system to consider it as valid information. As a result, the synthetic images can be then used as a part of an extensive dataset for pattern identification and other vital factors in the automatic diagnosis approach.
Other researchers have also examined ways of image synthesis using manipulated photographs. Due to the lack of databases that would be extensive enough to use in creating an effective machine-learning algorithm, Zhou et al. (2019) also suggest using existing photographs, manipulating them to create slight differences, and using them as new sources of information for deep learning. Thus, the researchers mention that rotating an image is not enough for the technique to be effective in terms of giving the system data it would deem new. However, the proposed image synthesis technique consists of using grading and creating lesions that would then be interpreted as disease progression. In this case, similarly to the previous one, fewer authentic images are needed since the model implies the use of several glaucoma pictures that would then be altered for creating new data. The technique has been referred to as beneficial by the authors of the study. The results show that the machine-learning algorithms detected the changes and analyzed them as they would be different pictures. Thus, more information was possible to use, which is always a positive factor when the results are affected mainly by the amount of information used in the process of comparing patterns and detecting problems in the structure of the eye.
Deep-machine learning relies on specific factors that ought to be examined in order for the results to show effective glaucoma detection. The most often used method of examining visual imagery is using a convolutional neural network (CNN). Multiple studies have resulted in this method being called effective for machine-learning glaucoma detection. For example, Li et al. (2020) propose an attention-based CNN for the identification of deteriorations based on fundus images. Thus, based on the proposed framework, the machine learning technique is to be centred around mimicking cognitive attention to the factor implied by the expert. The input data is selected in terms of the parts that are to be examined in order for the results to be positive. According to the authors of the studies, the network that is focused on smaller yet more important parts of data is more accurate in detecting issues compared to the system that implies the comparison of every bit of data available. This aligns with the previous argument, which suggests using the same number of pictures while minimizing the data by reducing dimension and feature extraction. Thus, this is another relevant study using this technique and illustrating its efficacy in detecting morphological issues while being more selective in terms of the amount of information that is being extensively analyzed. The use of CNNs was also found to be effective in the study done by Raghavendra et al. (2018). Based on the conclusions which have been resulted from the findings of the study, machine learning through the use of CNNs can proficiently detect glaucoma symptoms with very few deviations.
Besides the need for an extensive data basis, the algorithms have to be able to use the existing photographs effectively. This means that the characteristics of the image have to be maximized in order for the results to be as accurate as possible. A study on CNN for automatic glaucoma detection through fundus images suggests a unique technique used in the pre-processing procedure (de Moura Lima et al., 2018). The method involves using opposite colors in the images included in the research. Thus, pre-processing is based on the alteration of the image in order for the algorithms to be more sensitive to comparisons between normal eyes and glaucoma. Based on the results illustrated by the researchers, this method was effective in the detection of the condition, which suggests it is a beneficial image synthesis model that can be used for deep-learning on fundus images.
It is essential to point out that certain limitations may impact the accuracy of the results when it comes to using fundus imaging in machine learning. As mentioned prior, most studies show that fundus images lead to a more beneficial result in regards to effective diagnosis compared to other techniques. However, as suggested by relevant literature, images cannot always detect all the necessary factors that illustrate the presence of glaucoma. Gheisari et al. (2021) point out that glaucoma is often accompanied by blood flow deregulation, which is impossible to detect on images. Thus, this major symptom can be missed, which can become the reason the diagnosis is incorrect; hence, the detection of the condition will only be possible when the symptoms worsen and cause discomfort for the patient. As a result, fundus images may not be enough even when using them for machine learning algorithms, which have been shown to be extremely sensitive to glaucoma symptoms. However, this does not imply they are to be disregarded. Instead, fundus images are effective in measuring the cup-to-disc ratio and the corneal thickness (Liang, Zhang and He, 2018). These are often enough to detect changes in the optic nerve, which is ultimately what is damaged because of the presence of glaucoma. Moreover, the limitation can be confronted with the use of fundal videos. This is an excellent solution to the inability of pictures to show the blood flow deregulations while allowing the machine-learning technique to assess other symptoms related to the condition.
The need for further research in the field is highlighted by the problem itself. Glaucoma, as a primary cause of blindness, is difficult to detect, monitor, and address. Since the damage caused by the condition is often irreversible, the only solution left is diagnosing it in the early stages and addressing it. The diagnosis is especially helpful when the most severe symptoms are not yet detectable by the patient. Moreover, research in the field of machine learning glaucoma detection using fundus images is also vital because of an increase in surgeries related to the condition. Recent studies have shown that more individuals need surgical interventions due to the severity of their symptoms which are linked to the glaucoma diagnosis that has not been detected prior (Rathi et al., 2021). As mentioned prior, early detection is critical when it comes to avoiding the more prominent problems that correlate with optic nerve damage, including blindness. This, confronting the issue early on is not only useful in avoiding invasive procedures but also increases the possibility of beneficial outcomes for individuals. Automated detection learning has been extensively examined by researchers through studies of various optic disc and cup segmentation methods and retinal image synthesis techniques. As a result, studies show that the automated approach is not negatively impacted by subjectivity, unlike the non-machine diagnosis measures. Thus, relevant literature supports the use of machine learning in the examination of fundus images as a way of building an algorithm designed to detect glaucoma by comparing it with pictures of healthy eyes.
Reference List
Al-Bander, B., Williams, B., Al-Nuaimy, W., Al-Taee, M., Pratt, H. and Zheng, Y. (2018) ‘Dense fully convolutional segmentation of the optic disc and cup in colour fundus for glaucoma diagnosis’, Symmetry, 10(4), p. 87.
An, G., Omodaka, K., Hashimoto, K., Tsuda, S., Shiga, Y., Takada, N., Kikawa, T., Yokota, H., Akiba, M. and Nakazawa, T. (2019) ‘Glaucoma diagnosis with machine learning based on optical coherence tomography and color fundus images’, Journal of Healthcare Engineering, pp. 1–9.
Barros, D. M., Moura, J. C., Freire, C. R., Taleb, A. C., Valentim, R. A. and Morais, P. S. (2020) ‘Machine learning applied to retinal image processing for glaucoma detection: review and perspective’, BioMedical Engineering OnLine, 19(1).
Baudouin, C., Kolko, M., Melik-Parsadaniantz, S. and Messmer, E. M. (2021) ‘Inflammation in glaucoma: from the back to the front of the eye, and beyond’, Progress in Retinal and Eye Research, 83, p. 100916.
Carrillo, J., Bautista, L., Villamizar, J., Rueda, J., Sanchez, M. and rueda, D. (2019) ‘Glaucoma detection using fundus images of the eye’, 2019 XXII Symposium on Image, Signal Processing and Artificial Vision (STSIVA).
Castro, D. L., Valenti, C. and Tegolo, D. (2020) ‘Retinal image synthesis through the least action principle’, 2020 5th International Conference on Intelligent Informatics and Biomedical Sciences (ICIIBMS).
Civit-Masot, J., Dominguez-Morales, M. J., Vicente-Diaz, S. and Civit, A. (2020) ‘Dual machine-learning system to aid glaucoma diagnosis using disc and cup feature extraction’, IEEE Access, 8, pp. 127519–127529.
Clement, C. I., Howes, F., Ioannidis, A. S., Shiu, M. and Manning, D. (2019) ‘One-year outcomes following implantation of second-generation trabecular micro-bypass stents in conjunction with cataract surgery for various types of glaucoma or ocular hypertension: multicenter, multi-surgeon study’, Clinical Ophthalmology, Volume 13, pp. 491–499.
de Moura Lima, A. C., Maia, L. B., Pinheiro Pereira, R. M., Junior, G. B., Dallyson Sousa de Almeida, J. and de Paiva, A. C. (2018) ‘Glaucoma diagnosis over eye fundus image through deep features’, 2018 25th International Conference on Systems, Signals and Image Processing (IWSSIP).
Fatima Bokhari, S. T., Sharif, M., Yasmin, M. and Fernandes, S. L. (2017) ‘Fundus image segmentation and feature extraction for the detection of glaucoma: a new approach’, Current Medical Imaging Reviews, 14(1), pp. 77–87.
Gan, K., Liu, Y., Stagg, B., Rathi, S., Pasquale, L. R. and Damji, K. (2020) ‘Telemedicine for glaucoma: guidelines and recommendations’, Telemedicine and e-Health, 26(4), pp. 551–555.
Gheisari, S., Shariflou, S., Phu, J., Kennedy, P. J., Agar, A., Kalloniatis, M. and Golzan, S. M. (2021) ‘A combined convolutional and recurrent neural network for enhanced glaucoma detection’, Scientific Reports, 11(1).
Hark, L. A., Katz, L. J., Myers, J. S., Waisbourd, M., Johnson, D., Pizzi, L. T., Leiby, B. E., Fudemberg, S. J., Mantravadi, A. V., Henderer, J. D., Zhan, T., Molineaux, J., Doyle, V., Divers, M., Burns, C., Murchison, A. P., Reber, S., Resende, A., Bui, T. D., Lee, J., Crews, J. E., Saaddine, J. B., Lee, P. P., Pasquale, L. R. and Haller, J. A. (2017) ‘Philadelphia telemedicine glaucoma detection and follow-up study: methods and screening results’, American Journal of Ophthalmology, 181, pp. 114–124.
Hark, L. A., Myers, J. S., Rahmatnejad, K., Wang, Q., Zhan, T., Hegarty, S. E., Leiby, B. E., Udyaver, S., Waisbourd, M., Leite, S., Henderer, J. D., Pasquale, L. R., Lee, P. P., Haller, J. A. and Katz, L. J. (2018) ‘Philadelphia telemedicine glaucoma detection and follow-up study’, Journal of Glaucoma, 27(11), pp. 999–1008.
International Council of Ophthalmology (2017) ‘ICO guidelines for diabetic eye care’, International Council of Ophthalmology.
Jiang, Y., Tan, N. and Peng, T. (2019) ‘Optic disc and cup segmentation based on deep convolutional generative Adversarial Networks’, IEEE Access, 7, pp. 64483–64493.
Kausu, T. R., Gopi, V. P., Wahid, K. A., Doma, W. and Niwas, S. I. (2018) ‘Combination of clinical and multiresolution features for glaucoma detection and its classification using fundus images’, Biocybernetics and Biomedical Engineering, 38(2), pp. 329–341.
Khalil, T., Akram, M. U., Khalid, S. and Jameel, A. (2017) ‘An overview of automated glaucoma detection’, 2017 Computing Conference.
Khan, S. I., Choubey, S. B., Choubey, A., Bhatt, A., Naishadhkumar, P. V. and Basha, M. M. (2021) ‘Automated glaucoma detection from fundus images using wavelet-based denoising and machine learning’, Concurrent Engineering.
Kumar, N., Francesco, B. and Sharma, A. (2019) ‘Smartphone-based Gonio-imaging’, Journal of Glaucoma, 28(9).
Kwon, S., Kim, S. H., Khang, D. and Lee, J. Y. (2020) ‘Potential therapeutic usage of nanomedicine for glaucoma treatment’, International Journal of Nanomedicine, Volume 15, pp. 5745–5765.
Li, L., Xu, M., Liu, H., Li, Y., Wang, X., Jiang, L., Wang, Z., Fan, X. and Wang, N. (2020) ‘A large-scale database and a CNN model for attention-based glaucoma detection’, IEEE Transactions on Medical Imaging, 39(2), pp. 413–424.
Liang, L., Zhang, R. and He, LY (2018) ‘Corneal hysteresis and glaucoma’, International Ophthalmology, 39(8), pp. 1909–1916.
Lusthaus, J. and Goldberg, I. (2019) ‘Current management of glaucoma’, Medical Journal of Australia, 210(4), pp. 180–187.
Maetschke, S., Antony, B., Ishikawa, H., Wollstein, G., Schuman, J. and Garnavi, R. (2019) ‘A feature agnostic approach for glaucoma detection in Oct volumes’, PLOS ONE, 14(7).
Mitra, A., Banerjee, P. S., Roy, S., Roy, S. and Setua, S. K. (2018) ‘The region of interest localization for glaucoma analysis from retinal fundus image using deep learning’, Computer Methods and Programs in Biomedicine, 165, pp. 25–35.
Myers, J. S., Fudemberg, S. J. and Lee, D. (2018) ‘Evolution of optic nerve photography for glaucoma screening: a review’, Clinical & Experimental Ophthalmology, 46(2), pp. 169–176.
Prabhakar, B., Singh, R. K. and Yadav, K. S. (2021) ‘,Artificial Intelligence (AI) impacting diagnosis of glaucoma and understanding the regulatory aspects of AI-based software as medical device’, Computerized Medical Imaging and Graphics, 87, p. 101818.
Raghavendra, U., Fujita, H., Bhandary, S. V., Gudigar, A., Tan, J. H. and Acharya, U. R. (2018) ‘Deep convolution neural network for accurate diagnosis of glaucoma using digital fundus images’, Information Sciences, 441, pp. 41–49.
Rathi, S., Andrews, C. A., Greenfield, D. S. and Stein, J. D. (2021) ‘Trends in glaucoma surgeries performed by glaucoma subspecialists versus nonsubspecialists on Medicare beneficiaries from 2008 through 2016’, Ophthalmology, 128(1), pp. 30–38.
Rodriguez-Una, I. and Azuara-Blanco, A. (2019) ‘New technologies for glaucoma detection’, Asia-Pacific Journal of Ophthalmology.
Saba, T., Bokhari, S. T., Sharif, M., Yasmin, M. and Raza, M. (2018) ‘Fundus image classification methods for the detection of glaucoma: a review’, Microscopy Research and Technique, 81(10), pp. 1105–1121.
Sahlsten, J., Jaskari, J., Kivinen, J., Turunen, L., Jaanio, E., Hietala, K. and Kaski, K. (2019) ‘Deep learning fundus image analysis for diabetic retinopathy and macular edema grading’, Scientific Reports, 9(1).
Sarhan, A., Rokne, J. and Alhajj, R. (2019) ‘Glaucoma detection using image processing techniques: a literature review’, Computerized Medical Imaging and Graphics, 78, p. 101657.
Schuster, A. K., Erb, C., Hoffmann, E. M., Dietlein, T. and Pfeiffer, N. (2020) ‘The diagnosis and treatment of glaucoma’, Deutsches Ärzteblatt international.
Septiarini, A., Khairina, D. M., Kridalaksana, A. H. and Hamdani, H. (2018) ‘Automatic glaucoma detection method applying a statistical approach to fundus images’, Healthcare Informatics Research, 24(1), p. 53.
Serte, S. and Serener, A. (2019) ‘A generalized deep learning model for glaucoma detection’, 2019 3rd International Symposium on Multidisciplinary Studies and Innovative Technologies (ISMSIT).
Tan, N. Y. Q., Friedman, D. S., Stalmans, I., Ahmed, I. I. and Sng, C. C. A. (2020) ‘Glaucoma screening: where are we and where do we need to go?’, Current Opinion in Ophthalmology, 31(2), pp. 91–100.
Tanito, M., Sugihara, K., Hara, K. and Takai, Y. (2018) ‘Different glaucoma types and glaucoma surgeries among different age groups’, Graefe’s Archive for Clinical and Experimental Ophthalmology, 256(10), pp. 2013–2014.
Thakur, N. and Juneja, M. (2020) ‘Classification of glaucoma using hybrid features with machine learning approaches’, Biomedical Signal Processing and Control, 62, pp. 102-137.
Wang, S., Yu, L., Li, K., Yang, X., Fu, C.W. and Heng, P.A. (2019) ‘Boundary and entropy-driven adversarial learning for fundus image segmentation’, Lecture Notes in Computer Science, pp. 102–110.
Wu, C.W., Shen, H.L., Lu, C.J., Chen, S.H. and Chen, H.Y. (2021) ‘Comparison of different machine learning classifiers for glaucoma diagnosis based on spectralis Oct’ Diagnostics, 11(9), p. 1718.
Yadav, K. S., Rajpurohit, R. and Sharma, S. (2019) “’Glaucoma: current treatment and impact of advanced drug delivery systems’, Life Sciences, 221, pp. 362–376.
Yu, S., Xiao, D., Frost, S. and Kanagasingam, Y. (2019) ‘Robust optic disc and cup segmentation with deep learning for glaucoma detection’, Computerized Medical Imaging and Graphics, 74, pp. 61–71.
Zhou, Y., He, X., Cui, S., Zhu, F., Liu, L. and Shao, L. (2019) ‘High-resolution diabetic retinopathy image synthesis manipulated by grading and lesions’, Lecture Notes in Computer Science, pp. 505–513.