|Photo By Artemenko Daria via Shutterstock|
The early diagnosis of eye disorders is very important to determine the right course of treatment and the possible outcome of the patient. However, whatever time specialists could use to treat their patients gets used up in the diagnosis of these eye diseases. So, the researchers at the Singapore National Eye Center showed the potential of artificial intelligence in diagnosing eye problems.
In the study, the researchers included three major eye disorders – diabetic retinopathy, glaucoma, and age-related macular degeneration. Each disorder has its own mechanism that contributes to the number of cases of blindness in the world.
1. Diabetic retinopathy is a serious eye condition associated with diabetes mellitus, characterized by progressive damage on the eye’s retina. The mechanism of this eye disorder involves the excessive amount of sugar in the blood which damages various parts of the body including the eyes. Moreover, the excessive amount of sugar also makes it difficult for essential nutrients to be carried to the eyes. Over the course of time, diabetes causes the tiny blood vessels in the eye to leak blood and other fluids. The leakage causes the tissues of the retina to swell, resulting in blurry or cloudy vision. About one-third of the estimated 285 million people with diabetes may have signs of diabetic retinopathy.
2. Glaucoma is the most common cause of blindness in people aged 60 and older, and is the second leading cause of blindness globally. The eye condition happens when the fluid in the eyes does not drain properly. The accumulation of fluid in the eye causes pressure to build up. Eventually, the eye pressure can damage the optic nerve that leads to blindness. Symptoms of glaucoma include loss of side vision, pain in the eyes, seeing halos or colored rings, and blurred vision. Over 4 million Americans have glaucoma and half of them are aware of it. The suspected number of cases of glaucoma worldwide is 70 million.
3. Age-related macular degeneration affects the part of the eye responsible for seeing fine detail. It also affects the central vision that required for certain activities, such as reading, driving, and sewing. While the disorder does not cause pain, the blindness of the central vision can reduce the person’s functionality to finish daily tasks. According to the National Eye Institute, more than 1.75 million of people in the United States are affected by age-related macular degeneration. It is expected to increase by nearly 3 million by the end of 2020.
Treatment options for these eye disorders depend on their stages, and eye specialists need to properly diagnose patients to recommend the best treatment, in which an AI-powered diagnostic tool can help.
“With the AI system, results (for the screening) should be instantaneous and it can potentially reduce 80 percent of the workload of graders and optometrists, freeing up their time for treatment,” said Wong Tien Yin, the lead of the study and director of the Singapore National Eye Center.
The new AI system is designed to screen, diagnosis, and determine prognosis or risk stratification of patients, but not replace ophthalmologists, optometrists, physicians, and other eye specialists. The system acts as a complementing tool to lighten up the workload of the clinical staff and speed up the screening and diagnosing processes.
The researchers showed in their study that the AI system is equipped with a deep learning system, allowing it to learn new information. Their goal is to make the DLS screen eye diseases with sensitivity and specificity rates at par with standard equipment. To achieve this, they trained the DLS with 494,661 retinal images for detecting for diabetic retinopathy using 76,370 images, for possible glaucoma using 125,189 images, and for the age-related macular degeneration using 72,610 images. The other images are used to evaluate the performance of DLS in detecting the eye diseases. The training was completed in May 2016 while the validation of DLS was completed a year later.
The findings of the evaluation and validation revealed that the DLS has 90.5 percent sensitivity and 91.6 percent specificity for referable diabetic retinopathy. For the vision-threatening diabetic retinopathy, the DLS has 100 percent sensitivity and 91.1 percent specificity. In terms of detecting possible glaucoma, the DLS has 96.4 percent sensitivity and 87.2 percent specificity, while in age-related macular degeneration, the sensitivity is 96.4 percent and the specificity is 87.2 percent. The measurement of sensitivity and specificity has been referred to professional grades including retinal specialists, general ophthalmologists, and optometrists. The AI system and its DLS feature still need further research to be approved for healthcare settings.
“AI is used a lot when it comes to image analysis. That’s something that’s really exciting. AI is also helping us build predictive models of who is going to respond to a particular therapy,” said Dr. Olivier Elemento, the Director of the Caryl Israel Englander Institute for Precision Medicine at Weill Cornell Medicine.
[메디컬리포트=Ralph Chen 기자]