The second-leading cause of blindness globally, glaucoma is caused by increased intraocular pressure. Glaucoma affects the lives of nearly 3 million people in the United States. The key to preventing blindness in people with glaucoma is detecting the condition as early as possible.
For that reason, individuals should have regular eye screenings and undergo clinical tests in intervals determined by age and risk factors. These five tests help provide accurate diagnosis of glaucoma so treatment can begin, yet they take a great deal of time. This can delay treatment should the tests point toward glaucoma so researchers continue to search for faster ways to accurately detect the disease.
Top 4 Most Popular Ai Articles:
1. Basics of Neural Network
2. AI, Machine Learning, & Deep Learning Explained in Five Minutes
3. How To Choose Between Angular And React For Your Next Project
4. Are you using the term ‘AI’ incorrectly?
The search for faster detection recently inspired a group of scientists at New York University (NYU) to team with IBM researchers. Their goal is to determine how artificial intelligence (AI) could help both ophthalmologists and optometrists diagnose glaucoma more quickly.
An AI-Driven Solution to More Efficient Glaucoma Diagnosis
In current practice, five different clinical tests are used to detect glaucoma. These tests include tonometry, the measure of inner eye pressure; perimetry, which tests visual fields; ophthalmoscopy, looking directly at the optic nerve inside the eye; pachymetry, the measure of corneal thickness, and gonioscopy, which looks at the angle between the iris and cornea.
Currently, clinicians may use optical coherence tomography (OCT) imaging to supplement the five tests. Using the properties of light waves, OCT allows clinicians to take a cross-sectional image of the retina. Doing so helps practitioners visualize and quantify the retinal nerve fiber layer (RNFL), which is significantly altered over the course of glaucoma. However, this technique comes with limitations.
Much like the five tests, OCT imaging requires time to arrive at a diagnosis. Once an image is taken, clinicians typically have to clean up the input data. This minimizes the variability of the data and maximizes the performance of the test. For example, all images are reoriented to the same direction before analysis of the RNFL begins.
The NYU-IBM team developed a deep learning framework that helps detect glaucoma by utilizing raw images taken via OCT. By passing OCT images through the framework, the researchers achieved an accurate diagnosis 94 percent of the time. In addition, they were able to do so without running further tests or further “scrubbing” the data using traditional, time-consuming methods.
With this technique, a high level of accuracy is maintained when processing raw OCT images. As a result, the deep learning approach created by NYU and IBM streamlines the process and eliminates additional steps, which are time-consuming and can discourage clinicians from using the technology.
Comparing New Technology to Existing Methodology
The researchers examined a group of 624 people, about two-thirds of whom were glaucoma patients. As stated above, the deep learning methodology correctly detected cases of glaucoma 94 percent of the time. By comparison, the time-consuming, correction-based method led to a correct glaucoma diagnosis only 86 percent of the time.
The researchers pointed to two major factors to explain the increased accuracy. First, errors occurring during automated segmentation are no longer a concern since this processing is not necessary. Furthermore, the new methodology allows for the inclusion of image regions usually discarded during processing.
The NYU and IBM team behind the AI algorithm for detecting glaucoma went against the grain of current trends in AI development. Most often, AI development teams use large and deep networks to generate results. The individuals behind the new product understood the confidential nature of medical information means access is restricted.
So the researchers created a small, five-layer network, which is more efficient in the context of limited medical information. A larger structure may seem more robust at first, but would not actually add value. The smaller algorithm runs with great efficiency to achieve the initial goal of offering a less time-intensive means of diagnosing glaucoma.
AI’s Future in Ophthalmologic and Medical Diagnosis
The researchers behind the new algorithm are using AI in a number of other applications that are of interest to ophthalmologists and optometrists. For example, IBM Research and George & Matilda Eyecare recently announced they will explore deep learning AI to help clinicians diagnose eye diseases outside of glaucoma through imaging alone.
Deep learning is a subfield of machine learning inspired by the structure of the brain. The technology is used to create artificial neural networks. One of the most promising aspects of deep learning is called feature learning. This is the ability of artificial networks to perform automatic feature extraction from raw data, such as a medical image. Feature learning is what makes deep learning such a promising aspect of AI in the medical diagnostic industry.
George & Matilda Eyecare maintains a large database of anonymous clinical data, including imaging studies to be used in developing additional diagnostic algorithms. These new AI applications may speed the diagnosis of a wide range of conditions, get patients connected to treatment sooner and ultimately provide patients more time with intact vision. Many eye diseases, including glaucoma, are degenerative in nature, so fast and efficient diagnosis leads to much better outcomes.