11.12.2024

 

Understanding Vision Loss and the Need for Early Treatment

Researcher in Focus: Jesse Grootjen

What is your research about?

I’m not sure if you wear contacts or glasses, or have had them at some point in your life, if you had you probably someday realized you cannot read the subtitles of a movie anymore or see what’s on the slides during a lecture. At this point you would visit an optician and get your eyes checked and resolve your sight with glasses. You eventually might have found out you have been walking around with progressively worsening eyesight for months or even years. This is because you’re good at compensating for the loss of sight. In the case of myopia (nearsightedness; the example of needing glasses for seeing subtitles) you’ll end up squinting your eyes to be able to see things in the distance. Now for myopia, this is relatively easy to resolve. But for other visual impairments like cataracts (cloudy area in the lens of your eye), age-related macular degeneration (loss of central vision) or glaucoma (loss of peripheral vision) it is either permanent damage with regular treatment requirements or needs surgery to get back to normal vision. My research focuses on understanding how these compensation strategies look in terms of eye movements and whether machine learning can be used as an early detection tool. This approach could enable earlier treatments for individuals with visual impairments, helping to preserve as much of their vision as possible.

In what ways does machine learning enhance early detection of visual impairments?

The advantage of machine learning is that we can train models to recognize mitigation strategies we use to compensate for visual impairments really early, and deploy them in everyday devices. An example is that we could leverage the power of a webcam (or front-facing camera on your phone) in the future to recognize these compensations while you’re a device user and don’t get interrupted. We can analyze your eye movements while you’re reading or texting and give you a message if our model detects something is off. These types of early warnings are something that the Apple Watch already does with heart rate.

What challenges do you face in using machine learning to detect subtle changes in eye movements associated with early stages of visual impairment? How accurate are current models in identifying these early signs?

I believe we’re still far away from actually deploying these kinds of systems on everyday devices. For our current research, we are using high-quality and high-frequency eye trackers, and we’ve only been able to work with a couple of patients and visual simulations of visual impairments. Finding individuals at an early stage is difficult as most people are simply unaware at that stage. However, working with visual simulations of macular degeneration and training those without visual impairments to develop a preferred retinal locus (similar to one that could be developed as a mitigation strategy for someone with macular degeneration) has provided us with some interesting first results that highlight the potential. Moving from a lab setting in a highly controlled environment to webcams or front-facing cameras however opens many challenges, like changing lighting conditions, frequency, resolution, and many more.

How adaptable is your machine learning model to different visual impairments, like cataracts, glaucoma, or age-related macular degeneration? Are there unique challenges in creating models that can generalize across these conditions?

Currently, we are developing individual models for the different visual impairments. There are some unique challenges to creating a generalizable model encompassing all visual impairments, which is twofold. The first challenge is that visual impairments might have overlapping ways of manifesting. An example of this is the loss of visual acuity that is associated with cataracts, glaucoma, and myopia. The second challenge is even larger. Glaucoma for example generally consists of loss of visual acuity, loss of contrast, a shift in color, sensitivity to light, and depending on which version of glaucoma one might experience the loss of peripheral vision. Although we have interviewed many individuals who only mention being affected by a subset of these, where one person with Glaucoma might experience the loss of visual acuity, loss of visual acuity, and sensitivity to light, the next person could have neither but will experience the color shift and has loss of peripheral vision.

What advice would you give to others considering doing their PhD/PostDoc in Munich?

I think there is lots of advice to give, but most importantly, make sure you like the people in the group you’re considering to join and you like the topic. In the end you’ll end up spending 3-5 years of your life there and spending it on a topic you like makes the whole PhD a lot easier.

Read more about the study mentioned by Jesse Grootjen

Jesse Grootjen, Alexandra Sipatchin, Siegfried Wahl, Tonja-Katrin Machulla, Lewis Chuang, and Thomas Kosch. 2023. Assessing Eye Tracking for Continuous Central Field Loss Monitoring. In Proceedings of the 22nd International Conference on Mobile and Ubiquitous Multimedia (MUM ‘23). Association for Computing Machinery, New York, NY, USA, 54–64. DOI

11.12.2024