Into the Deep

Artificial intelligence (AI) has transformed ophthalmology over the last decade, enabling the analysis of all types of data and images, and achieving convincing performance in areas such as the segmentation of retinal layers and the classification of eye diseases. Deep learning, a relatively juvenile subfield within AI, has the potential to expand on what can be achieved with machine learning algorithms in healthcare – assisting in the identification of patients for clinical studies, and facilitating the screening, diagnosis and management of major eye diseases. RetinAI CEO, Carlos Ciller, tells us how this technology can optimize clinical workflows and connect ophthalmologists to whole networks of valuable information.

What are the key challenges that ophthalmologists currently face?

Today, many healthcare institutions are forced to navigate a sea of unstructured and disorganized patient data. From a clinical and research standpoint, this lack of structure limits the extent to which organizations can develop new drugs, technologies and medical devices. That same lack of structure also adds expensive operational layers to other tasks, such as patient identification for clinical trials and the analysis of that patient after they have enrolled.

Similarly, hospitals are under increasing pressure to care for an aging population that is growing much faster than the number of healthcare professionals needed to support them. Today, an ophthalmologist may be required to assess multiple visits from a diabetes or AMD patient over a number of years and make multiple medical imaging acquisitions for each visit. Findings must be quickly retrieved and evaluated to cope with the time constraints of the clinical setup – which puts a lot of stress on the clinician.

These are both areas where data-driven insights, process automation and connectivity would provide a deeper and more meaningful understanding about the patient’s condition from the very beginning. Any assistance provided to healthcare professionals for clinical decision support (CDS) would simplify and speed up the clinical day-to-day practice, while simultaneously increasing the number of patients seen at the clinic.

How is RetinAI using AI to help ophthalmologists overcome these challenges?

At RetinAI, we focus on optimizing and simplifying clinical, research and pharmaceutical workflows, bringing efficiency back to healthcare institutions.

We started in an ophthalmic research lab in Switzerland, creating standalone software for processing medical eye scans using AI. Initially we worked on large-scale analysis of medical images used in clinical trials, but after a few months, we realized that having some of the best AI algorithms in the market wasn’t enough. If we wanted to make a difference, we needed to build a software link between the machines, the patients and the physicians – and provide a fully integrated overview of the clinic.

Today, we are developing a ground-breaking software platform that can be used for clinical and pharmaceutical research, and for CDS. RetinAI Discovery® enables the analysis and transfer of ocular medical data, connecting devices, data sources and decisions in a seamless and easy manner. Created under the umbrella of the ISO and MDR standards, the platform works on commodity hardware available via a hospital’s web browser, which we believe will enable it to become part of routine medical care in the future.

What benefits does RetinAI Discovery offer clinicians?

RetinAI Discovery assists both the researcher and the physician throughout the course of the disease, enabling the automatic identification of patient candidates for clinical trials with an expert level of precision. By using Discovery’s image processing technology, doctors can extract information from patient history, detect patients that would fit a specific search criterion and cross-check their eligibility through automatic evaluation of OCT or information extracted through EHRs. This digital tool also makes it easier to interact with fellow colleagues, and connects users to a valuable network of knowledge from day one (see Figure 1).

So far, the RetinAI Discovery® platform has supported the evaluation of data obtained during clinical trials from different manufacturers across the globe, enabling researchers to gain an insight into the application of a drug and its impact on specific biomarkers over time. More specifically, it has quantified OCT image biomarker evolution (retinal layer thickness, intraretinal and subretinal fluid, pigment epithelium detachment and geographic atrophy) for conditions like AMD and diabetic retinopathy. We have also developed a unique proprietary technology that enables thorough validation of ocular images – even before processing them with our AI modules – as we are acutely aware of the perceived risks associated with using AI algorithms. 

Other potential uses of RetinAI Discovery for clinics and hospitals are fostering interoperability, integration and the structuring of data by design. By transforming multiple medical inputs into a unified and device-agnostic standard format, we are able to aggregate information into a single workspace. And that’s especially relevant for clinics using multiple devices and vendors over a long period of time.

How do you see AI evolving in the future?

AI algorithm designs (primarily focused on convolutional neural network architectures), training methods, and almost all of the current AI strategies will remain the same for the next few years – what will change significantly is the availability of these algorithms and their validation.

More researchers will learn how to adapt current AI tools to reach their full potential and will understand that the real challenge is no longer in the algorithm – it is in the way the algorithm was generated, the quality of the annotations that were used to create it, and the availability of that algorithm to the rest of the healthcare community. Organization of knowledge will become the cornerstone of medical device manufacturers, research institutions and pharmaceutical companies alike. Multimodal, large-scale data curation, structuring and organization will be the next focus to enable multimodal AI analysis, and regulations will adapt to meet the requirements of patients.

We are confident that tools such as RetinAI Discovery are going to unlock a completely new type of collaborative research. As an ophthalmologist, you will be able to access all of your research faster and make it available to other professionals through the click of a browser. By breaking the technological barriers that separate clinicians from the best technology available, we will be able to deliver new findings and insights from existing information.