Rotterdam Study / ERGO
As of 1989, ophthalmology has been part of the Rotterdam Study. Retinal images of up to 8 study visits with a 4 to 5-year interval have been collected thus far. The images are being used for research focused on AMD, myopia and glaucoma. Elucidation of retinal features predictive for disease progression is one of our core interests.
We also study the contribution of lifestyle, genetics and the metabolome to disease incidence and the natural course of these diseases.
Retinal image analysis
During eye exams, retinal images are often taken with multiple devices in order to get a complete assessment of the eye. These images are then viewed and interpreted by the clinician, or by a team of graders in a research setting. EyeNED has a team of experienced graders who can grade images using WARMGS, MetaPM, and other grading standards. Our core expertise is grading of AMD, DR, myopia and glaucoma. However, due to our experience with grading of the Rotterdam Study, our graders are familiar with a wide variety of retinal conditions.
Artificial intelligence driven image analysis
Grading of retinal images is currently very laborious and is a skill that requires months to years to master. This, because retinal lesions are often subtle and difficult to distinguish. On top of this, the number of images that need to be analyzed will increase in the coming year, as imaging devices are now also present in many optician stores and the number of images taken at the clinic increases as well.
Digital solutions are revolutionizing the entire medical field and show great promise in disease classification and clinical decision-making. Artificial intelligence (AI) is mastering the field of retinal imaging, and deep learning has already provided many algorithms for diagnosing eye diseases and referral recommendations. We aim to improve these algorithms in close collaboration with other research teams around the world.
Platform for image segmentation/annotation
The performance of algorithms is highly dependent on the quality of the training data; the so called “ground truth”. Because of this, we are focused on developing high quality data for algorithm development. Human graders are annotating/segmenting all lesion in multi-modal images.
To aid the graders, we have developed a platform for multi-modal retinal image annotation. The multi modal images are all linked to each other, so the grader can see the exact location on multiple modalities.
Platform for image analysis and diagnosis
In the clinic, ophthalmologist currently need to analyze many different image modalities and time points to get an overview of the condition of the retina. Because of this, smaller lesions and subtle changes in time are easily missed. In order to help clinicians get better insight in the status of the retina, we are developing software that integrates all relevant clinical data. This will ultimately help the clinician in treatment decisions and supports the dialogue with patients about treatment options and outcomes.
Many algorithms that can help with the image analysis have been developed, but most professionals in the eye field do not have access to these algorithms, or can only use them in or in combination with a specific imaging device. There is a clear need for software platforms which run independently of devices and vendors, that can connect images from various sources and time points, and that allow integration of multiple AI solutions.
Using the retina to detect dementia, the EYE2BRAIN project
Alzheimer’s disease is rated among the worst conditions that can happen to you in life. In our aging population, the incidence of Alzheimer’s disease is increasing. This has a large impact on those affected but also on society and healthcare costs. Although medications to alter the course of Alzheimer are still under development, it is clear that a favorable lifestyle has a huge impact on the outcome of the disease.
Early detection and risk prediction can help create a turning point for lifestyle, and stimulate behavioral changes in individuals at risk to alter their fate. Unfortunately, cheap and reliable screening tools to detect risk of Alzheimer are still scarce.
Retinal imaging such as color fundus photography and optical coherence tomography (OCT) may offer a solution, as they reveal neural and vascular tissue in a non-invasive manner, and permit identification of abnormalities which are indicative of brain disease. Previous studies have proposed a link between retinal biomarkers and risk of various neurodegenerative diseases in the brain, but classical methods have not been powerful enough for prediction of these disorders on a patient level. This gap may be bridged by current technological advances in artificial intelgence (AI). AI by deep learning involves systems that can learn to find patterns in retinal images that are typical for certain conditions, but which may have been overlooked by the human eye. These features may be detectable years before first symptoms of neurological disease appear.