AI to Interpret
The future of eye care will rely on artificial intelligence-driven imaging interpretation – in the clinic and out
Andrey Kuropyatnyk, Maria Znamenska |
sponsored by Altris
AI has enormous potential: it can process information and images faster than humans and use data more effectively. No surprise then that AI is becoming integral to healthcare practices worldwide. Today, diagnosis, treatment and monitoring of many ocular conditions rely on ophthalmic imaging – in fact, disease screening, teaching, clinical trials and telemedicine often seem impossible without it. Andrey Kuropyatnyk, Chief Executive Officer and Maria Znamenska, Chief Medical Officer at Altris – a company responsible for many of these latest imaging technologies – explain how AI-driven OCT interpretation will continue to benefit ophthalmology practices in the coming year.
How are advances in AI changing the landscape of ophthalmology?
AK: Though the most significant change concerns the interpretation of fundus images and OCT scans, there are many areas that could leverage the potential of AI – including the prediction of disease progression and treatment outcomes, as well as neurologic and cardiovascular pathologies. More generally, AI may help improve patient flow, and assist doctors in capturing photographs and scans more accurately.
How is Altris using AI and machine learning to benefit ophthalmologists – and their patients?
MZ: Our specialty lies in OCT analysis. Devices are becoming more and more affordable, with almost all manufacturers producing top-level and general options. In optometry, OCT availability sets a practice apart from its competitors. Thanks to current software support, it is relatively easy to perform a comprehensive eye health OCT examination. The most important element for an optometrist is to understand if there are problems with the B-scan, especially in early disease stages. And that’s where Altris can help – we train AI to detect almost all possible diseases that can be seen on macular OCT scans. Altris AI software is capable of analyzing 128 OCT B-scans in 90 seconds or less and scores each for severity of pathological signs, regardless of the pathology. It allows optometrists to see any possible pathology and refer the patient to an ophthalmologist if needed.
We have a great case to illustrate this: there was an OCT macula examination containing 96 B-scans of a 38-year-old male. He had no complaints and a BCVA 20/20. While 95 B-scans were marked as “No pathology detected”, one from the parafoveal zone was identified as “Low severity” – our AI had detected the early stage of hard drusen, which most likely would have been missed in regular practice. In such cases, AI helps to determine whether the patient is at risk of AMD development, potentially saving their sight.
Altris AI software provides detailed analysis of almost 50 pathological signs and diagnosis recommendations for almost 40 diseases – including detection and visual segmentation, as well as urgency classification. Our software drastically improves the diagnostic process, ensuring patients leave the clinic with a true understanding of their condition, and allowing visiting retinal specialists to prioritize critical cases and focus only on pathological conditions. In short, AI makes eye care better by producing more accurate and faster diagnoses, making diagnostic process more efficient and cost-effective.
Who can benefit from AI and machine learning?
MZ: In our opinion, everyone. From technicians performing screening examinations to optometrists, general ophthalmologists and retina specialists. It can also be used as a great educational solution for medical students and residents, as it is capable of illustrating different pathologies and giving diagnosis recommendation. Though the focus seems to be on small clinical practices, they are not the only ones who can benefit from AI.
We have examples of large chains and reading centers who have leveraged AI’s potential. In one network of ophthalmology clinics, OCT examinations are performed by technicians – upwards of 4,000 per month. Our software detects pathological cases so those patients can be referred to retina specialists as soon as possible. In reading centers, Altris AI helps speed up triage, separating normal and pathological scans, and saving a lot of time for medical imaging specialists in the process.
Altris AI software can identify intraretinal and subretinal fluid, subretinal hyperreflective material, pigment epithelium and neurosensory retina detachments, intraretinal hyperreflective foci and hard exudates – the most common signs of wet AMD and diabetic retinopathy. In those pathologies, the most common prescription is anti-VEGF therapy. When to stop with injections is a common question of ophthalmologists – and one where AI can help. Altris AI algorithms can track changes in those pathological signs during follow-up examinations after each injection. Visualization and quantitative analysis of changes will help both doctors and patients to evaluate treatment results, increase compliance and consequently improve therapy.
How do you see AI and machine learning evolving in the near – and longer-term – future?
AK: Although many AI algorithms show great accuracy in the research phase, they prove weak in the clinic, despite being trained on a consistent, high-quality dataset. If we want to access the full potential of AI, we need to start bringing those lab results to reality by addressing the factors which influence real-world performance, such as low signal strength, eye motion and opacities. We must also solve a number of other problems, from overall AI performance to the security and protection of sensitive data. Cloud computational resources are effective but for many clinics, storing patients’ data outside of a clinic is impossible, as is data transfer to another location. Ethical issues must also be addressed, as they can affect the company brand and reputation, as well as the lives of employees, customers and shareholders.
There is talk of AI replacing doctors but that is not the case: AI is meant to complement daily practice. With AI systems, a clinic can take on more patients, patients can have more accessible care and earlier diagnosis, and doctors can save valuable time. AI has tremendous potential to help manage patients in a world of limited resources. On a practical level, doctors and AI systems make different kinds of mistakes so, if we are to achieve the highest accuracy, we need to get the best of both.
Though AI offers almost limitless opportunities in the interpretation of OCT scans, that does not mean there are no limitations. In cases where AI has been trained to detect commonly observed diseases on OCT scans, such as diabetic retinopathy or age-related macular degeneration, it may not bring clinical value to a patient with a macular hole, for example, because it would be diagnosed incorrectly. Misdiagnosis diminishes the benefits of AI, which is why it makes sense to only screen certain patients for certain pathologies.
In the long term, I think we will see more AI systems for disease prediction and treatment support, as well as robotic screening with OCT and fundus camera imaging in kiosks or primary care offices. In my mind, the benefits of AI far outweigh the risks, and I have no doubt it will become part of the future of healthcare.