Thumbnail website day 2__Artboard 4

Deep Learning in Focus at ESCRS 2024: The Future of Anterior Segment Ophthalmology

AI has arrived… sort of. Innovations in machine learning are the talk of the town, but obstacles to their implementation are clogging up the works

A Day 2 session at the 42nd Congress of the European Society of Cataract and Refractive Surgeons (ESCRS 2024) in Barcelona – a Clinical Research Symposia titled, Bringing Deep Learning to the Anterior Segment, featured pioneering presentations on the future of artificial intelligence in ophthalmology. The session highlighted the transformative impact of deep learning on diagnosing anterior segment disorders and explored federated learning’s role in revolutionizing big data management in eye care.

Federated learning: Big data from smaller datasets? 

In an era where data is king, the challenge of creating large, diverse datasets for ophthalmic research is more critical—and complicated—than ever. At ESCRS 2024, data scientist Dr. Michelle Hribar (USA) took the stage to unpack the complexities of federated learning, a concept that might just revolutionize how we think about big data in eye care.

Dr. Hribar kicked off by highlighting the importance of large datasets in reducing bias and improving accuracy in research. “Large, diverse datasets are the secret sauce to accurate models,” she said, adding that the traditional method of pooling data from multiple sites, known as registries, has been the go-to strategy for years.

While this method offers the advantage of centralized data access, it comes with hefty baggage—legal, privacy and resource-related challenges that make it less scalable. “Not to mention,” Dr. Hribar added, “sending large images can be like trying to fit an elephant through a keyhole.”

Enter the federated data network—Dr. Hribar’s proposed upgrade to the old system. This approach allows data to stay at its original site while still being accessible for analysis. The benefits? Scalability, broader data access and no more haggling over data-sharing agreements.

Dr. Hribar concluded by underscoring the need for large, diverse datasets in ophthalmic research. “Federated learning could be the game-changer we’ve been waiting for,” she said. She encouraged all researchers to join this effort, adding, “After all, who wouldn’t want to be part of something this revolutionary?”

Unlocking the potential of deep learning in eyecare

Artificial intelligence (AI) is no longer just a buzzword—it’s transforming healthcare, particularly in medical imaging. At the ESCRS 2024 congress, Mr. Sunny Virmani (USA) took the stage to demystify deep learning, a powerful subset of AI and its implications for ophthalmology.

Mr. Virmani kicked off by highlighting AI’s versatility, saying, “AI has great potential for many healthcare challenges, and the best part is that it serves the needs of many professionals, including business owners.” He then dove into the mechanics of AI, noting, “Deep learning, a special category of AI, has become particularly popular in the last few years, especially for its applications in medical imaging.”

When it comes to teaching AI how to recognize patterns in medical images, traditional programming requires specific instructions. But as Mr. Virmani explained, deep learning operates differently: “With supervised learning, you give the model examples and let it figure out how to make decisions. No need to spoon-feed every step.”

AI: A new frontier for early detection

Mr. Virmani then touched on AI’s role in early disease detection using external eye photos. “AI can detect health biomarkers in external eye images, potentially useful as a first-line at-home screening tool,” he explained, noting that early results are promising in making this technology more accessible through smartphone images.

However, AI isn’t just about building models—it’s about ensuring they work in real-world settings. “You can build these models, but if you don’t test their effectiveness in the hands of doctors and nurses, they can’t be useful to patients,” Mr. Virmani emphasized. The ultimate goal? Bringing these technologies closer to patients to revolutionize healthcare delivery.

For example, keratitis, a leading cause of blindness worldwide, isn’t packing its bags anytime soon. Dr. Nino Hirnschall (Austria) kicked off his presentation by highlighting the scale of the problem. With over 2 million people blinded by keratitis annually, it’s clear this isn’t just a small bump in ophthalmology’s road, he noted.

But there’s a silver lining—or rather, a digital one. AI is stepping up to streamline keratitis diagnosis. “AI can help differentiate infectious from non-infectious keratitis and even predict the best treatment approaches,” Dr. Hirnschall explained. While promising, there’s still a mountain to climb in terms of research: 

Dr. Hirnschall wrapped up with a call to action. “Pathogen detection with deep learning approaches is very promising, but more research on OCT scans is needed.” He also stressed the need for a European global database for keratitis images, adding, “We need a European global database for keratitis OCT and slit lamp images.”

Cracking keratoconus: Deep learning’s new eye on early detection

Keratoconus (KC) diagnosis has seen incredible advancements thanks to modern corneal topography, but the early stages of this progressive eye disease remain tricky to detect. At the ESCRS 2024 conference, Dr. Jorge Luis Alio Del Barrio (Spain) tackled the topic of how deep learning could be the game-changer for identifying keratoconus earlier and with more precision.

“Thanks to modern corneal topography, diagnosing established keratoconus has become relatively easy,” Dr. Del Barrio noted, adding with a grin, “However, catching it in the early stages? That’s still a bit like finding a needle in a haystack.” Early diagnosis, he stressed, often relies on a clinician’s experience and intuition—skills that not every laser vision correction practitioner may possess. “We need to support clinicians with automatic methods for detecting KC, and this is where deep learning steps in,” he emphasized.

Dr. Del Barrio explained that artificial neural networks (ANNs) and deep learning are branches of artificial intelligence (AI) that can mimic the brain’s ability to recognize patterns. “These algorithms can see what even a skilled observer might miss,” he said. In ophthalmology, deep learning has already proven effective in automating keratoconus diagnosis by analyzing data from Placido-based topography.

In conclusion, Dr. Del Barrio reinforced the potential of deep learning algorithms in keratoconus diagnosis. “They’ve proven their worth in detecting established cases, but we still have work to do in identifying early disease,” he said. He pointed out that future improvements will likely involve combining topography, OCT-based thickness maps, and biomechanical data, particularly for catching those elusive early cases.

DICOM demystified: Why it matters for ophthalmology

Today, digital images play a vital role in diagnostics, and understanding DICOM is crucial for integrating medical devices and systems in a seamless, standardized way. DICOM isn’t just for IT experts—it’s for every ophthalmologist aiming to streamline diagnostics and improve care.

Dr. Flora Lum (USA) addressed this topic by breaking down the complex world of DICOM for beginners, showing why it’s more than just tech jargon—it’s essential for modern ophthalmology. 

Dr. Lum painted an ideal future for ophthalmologists: “Imagine a world where you go to your workstation and get all the images you need for diagnosis and follow-up in one place. That’s why DICOM matters for ophthalmology.” DICOM, short for “Device Imaging and Communication Standards in Medicine,” was created in 1993 to handle the increasing flood of digital images in healthcare. She likened it to the Bluetooth and Wi-Fi standards we rely on in everyday life: “We’re getting there when it comes to medical device interoperability with DICOM.”

As Dr. Lum wrapped up her talk, she left a powerful message for ophthalmologists: “Demand DICOM standard images from your vendors. DICOM ensures that clinicians have access to more accurate and complete information, leading to better patient outcomes. So, let’s make that ideal world a reality.”

Finishing strong with foundation models

Artificial intelligence (AI) is reshaping healthcare, and ophthalmology is no exception. At ESCRS 2024, Prof. Pearse Keane (Ireland) shared how foundation models and transfer learning are driving this change, helping AI achieve more with fewer data.

With AI advancing at lightning speed, Prof. Keane’s presentation showcased how foundation models could revolutionize ophthalmology, making sophisticated AI tools more accessible and adaptable across various medical applications.

“What exactly are foundation models? Basically, they’re the latest buzzword for transfer learning—big models trained on massive amounts of unlabelled data, ready to be fine-tuned for specific tasks.” These models, according to him, are powerful because they don’t need expert labels to get started. “Self-supervised learning is the key here; the models essentially learn on their own.”

Prof. Keane highlighted the shift in AI architecture, moving away from traditional convolutional neural networks (CNNs) to transformer models. “Transformers are the new stars of the AI show. They handle different data types and massive quantities of it—this is the tech behind ChatGPT!” he explained.

One of the most exciting applications Prof. Keane discussed was vision-language models. “Imagine uploading an anterior segment image and asking, ‘Is this normal?’ or even requesting a diagnosis,” he said, painting a picture of AI-powered diagnostics that could become a reality sooner than we think.

With AI, the future always blends into the present at the speed of innovation. Prof. Keane didn’t shy away from the reality with a revelation of his own for RETFound: a 3D version of RETFound. “We’re scaling up from 2 million to 20 million images, exploring synthetic data, and even expanding into anterior segment applications,” he said, hinting at even more exciting developments on the horizon.

Editor’s Note: Reporting for this story took place at the 42nd Congress of the European Society of Cataract and Refractive Surgery (ESCRS 2024), held from 6-10 September in Barcelona, Spain.

Subscribe
Notify of
guest
0 Comments
Oldest
Newest Most Voted
Inline Feedbacks
View all comments