Sign up for our newsletter and get the latest HPC news and analysis.
Send me information from insideHPC:


IBM Research Applies Deep Learning for Detecting Glaucoma

Rahil Garnavi is a Research Manager, Research Staff Member, and Master Inventor at IBM Research.

Over at the IBM Blog, Rahil Garnavi writes that IBM researchers have developed new techniques in deep learning that could help unlock earlier glaucoma detection.

Earlier detection of glaucoma is critical to slowing its progression in individuals and its rise across our global population. Using deep learning to uncover valuable information in non-invasive, standard retina imaging could lay the groundwork for new and much more rapid glaucoma testing.

Developed in collaboration with New York University, the study shows how deep learning models can be trained to learn from and analyze easily-obtained retina images, and then use this analysis to directly estimate visual function – a critical indicator of glaucoma.

Glaucoma is the second leading cause of blindness in the world, affecting almost 4 percent of the world’s population aged 40 years or older. In 2010, 60.5 million individuals were affected by the disease, with these numbers expected to rise to 80 million by 2020. As a disease, it progresses slowly, and it is notoriously asymptomatic. Because of this, up to 40 percent of vision can often be lost before an individual notices. While treatments exist that help to prevent its progression, nothing can be done to restore vision itself. AI research like this detect glaucoma with imaging and deep learning analysis lays the groundwork to help alleviate one of its most disabling symptoms: blindness.

Visual field tests map how well patients see throughout the visual space, and are used to diagnose a variety of conditions. For example, optic nerve damage caused by glaucoma causes characteristic visual field defects in the upper and lower fields of view. While other conditions can affect retinal structures in a similar manner to glaucoma, the impact on vision is often very different. Thus, these tests are an integral part of the diagnostic process. However, these tests rely exclusively on patient feedback and thus, are subjective to the alertness of the patients. Time of day is known to be a factor that influences patients’ performance on these tests, where mornings are better than right after lunch. As a result, one may need multiple tests to obtain an accurate measurement of any vision loss.

From a biological point of view, we know there are associations between visual function and retinal structure. Here an interesting research question emerges: can we estimate visual function directly from structures in the eye that can be imaged using non-invasive techniques? The answer is yes, as we have discovered that there is information in retina imaging data that can help to assess the presence of glaucoma.

IBM Research, in collaboration with New York University, has conducted a study to explore this question, and employed a data-driven approach using deep learning techniques. Our study uses 3D raw OCT (Optical Coherence Tomography images of retina) imaging data to estimate corresponding visual field index (VFI) values with unprecedented accuracy, with an overall error within 2%. VFI is a global metric that represents the entire visual field, and accurately capturing that with AI offers to lay the groundwork for future technologies that can potentially use this analysis to quickly estimate a patient’s visual function. This could give professionals access to precise information – without the need for multiple and time-intensive tests – when gathering data for a glaucoma diagnosis.

Conventional OCT structural measurements, such as retinal nerve fiber layer (RNFL) thickness and ganglion cell inner plexiform layer (GCIPL) thickness, could not achieve this degree of accuracy, despite both layers being known target locations of glaucoma. Our study suggests the structural information captured by OCT contains information that is highly correlated with functional measurements, and could be extremely useful to professionals as they look to make a diagnosis.

The study was presented this week at the annual meeting of ARVO (the Association for Research in Vision and Ophthalmology).

Sign up for our insideHPC Newsletter

 

Leave a Comment

*

Resource Links: