In May 2024, the FDA granted de novo authorization to Notal Vision’s patient-operated Scanly Home OCT device, which operates in conjunction with the company’s AI-based Notal OCT Analyzer (NOA) to automatically grade self-administered images. The FDA clearance marked the first-ever approval of an AI algorithm for OCT images.
We spoke to Eric Schneider, a co-author of the report that describes the pivotal trial and published in Ophthalmology Science, which helped achieve the de novo authorization, and Nancy Holekamp, former Director of Retina Services at the Pepose Vision Institute and co-author of a study published in RETINA evaluating how Notal’s Home OCT system impacts patient treatment, to learn more about the devices.
How do the Notal OCT Analyzer (NOA) and the Home OCT (HOCT) system work together to monitor neovascular age-related macular degeneration (nAMD) progression?
Eric Schneider: The Home OCT system has two main components: the actual table-top imaging device, and the associated AI-based segmentation algorithm. This is a deep learning algorithm that segments “hyporeflective spaces,” which are clinically equivalent to intra- and subretinal fluid. We used the term “hyporeflective spaces” as opposed to “fluid” because not all fluid is hyporeflective on OCT, and there are some hyporeflective spaces that don’t represent exudative fluid, and so we had to be more specific about what was being segmented. The AI algorithm segments these spaces across all the individual B scans within a cube and calculates their volume, plots an enface map of where those hyporeflective spaces are located, and provides a trajectory to show whether that volume is increasing or decreasing.
Nancy Holekamp: AI algorithms are now pretty standard. It's very common for universities, researchers, industry, etc., to create their own AI algorithms. But Notal Vision was among the first. It's absolutely critical that we have an AI algorithm, because – if you have even a handful of patients doing this every day on both eyes – the throughput of images is so enormous that no human reader could keep up. It has to be read by AI, otherwise it's not a practical solution.
This is the first time it was done, and the FDA process was difficult because it's always difficult to be first. Regarding Eric’s point about “hyporeflective spaces,” it’s really interesting – the FDA resisted conversations about fluid because no one proved to the FDA that we were actually measuring fluid. So the language had to change.
What key metrics did you use to evaluate NOA’s performance in processing these home-based OCT scans?
Holekamp: One thing we had to agree on is repeatability – if you take the same test on the same OCT with the same patient, will you get the same answer every time? And will that answer match an in-office OCT on that patient? In addition, you have to show that the AI algorithm reads an OCT as well as a human grader, and show that the Home OCT captures the pathology as well as an in-office OCT.
Schneider: We also looked at how capable patients were at using the device – could they set up and calibrate the device and self-image effectively? And we found that a really high rate of people were able to self-image – somewhere between 93 and 96 percent – so this was another important metric.
How did the NOA system compare with human graders?
Holekamp: I think that in the long-term this technology will actually be better than human graders. Human graders have to be trained, they get tired, they go on vacation; they might even leave their job and then you have to train new people. But this technology gives reading center grading quality at every analysis of the OCT – every single time it brings expert analysis to retina specialists. That’s the beauty of it.
Given the increasing impact of AI, what advancements do you foresee for home-based retinal disease monitoring?
Schneider: Again, speaking in the long term, if we build up bigger quantities of data, really granular data, and train algorithms on this specific data, that’s going to be really valuable. That’s really how to best develop AI – the more data you have, the finer the algorithm and the better the insights you can gather. If you look at the literature, in terms of visual outcomes, there is a big mismatch regarding what we see in clinical trials and how we perform in the real world. I think this technology represents a great way of figuring out what's happening in the real world.
What are the biggest barriers to widespread adoption of home OCT, and how might they be addressed?
Schneider: Inertia is a big barrier to adoption. People are very used to doing things the way they have always been done. But when you consider what insights we can glean from looking at longitudinal volume data on these actively treated patients, you begin to gain insights you would miss looking at only infrequent in-office OCT.
Holekamp: Doctors have to be willing to trust the technology and realize that they don’t have to see their patient frequently if their patient is using the Home OCT at home. The FDA approval really helps with this; the FDA stamp of approval is highly regarded in the US. But it is difficult to change habits and our current practice is to monitor patients in the office, and now we are now going to substitute that for in-home monitoring on a near daily basis.
We will also have to make room in our daily practice to deal with emergency patients when an alert is triggered by the system. Right now, retina specialists have very busy offices where it's very hard to get in on a moment's notice. But this treatment paradigm is designed so that when an alert is triggered, the doctor’s practice can absorb these emergency office visits. This requires a paradigm shift in how retina specialists practice throughout the US.
Schneider: Another major barrier will be financial. A lot of retinal specialists are concerned about the impact home OCT will have on the value of in-office diagnostics, but many of those same providers don’t realize the added value of new codes in our specialty. New codes allow for additional reimbursement via coding for both in-office OCT and home monitoring. While budget neutrality exists, the budget is fixed for medicine as a whole, not retina specifically. Thus, more codes allow retina to have a larger piece of the budgetary pie. There's also a concern about increased physician workload, but having worked with the Home OCT Physician Portal for this device quite a bit, I find the review process very efficient and streamlined.
Is there anything else you would like to add?
Holekamp: I’ve used the system to manage many patients, and have published a study with Jeffrey Heier and other colleagues which details our experiences. The system is achieving all the things we aspire to when we use the term “personalized health care.” And I think it's better for patients. When you start using daily information, you realize that we’ve been managing patients with a paucity of data points.
Schneider: There's a lot of potential for this technology. Data really is king in medicine these days, and this is a great way to obtain a ton of data, which not only impacts patient care in the now, but could really help us generate a lot of data to get more insights into disease and train future AI algorithms on ways to better improve patient outcomes.
Holekamp: Using the Home OCT and seeing how many data points there are, we're learning more about the disease and we're managing patients better. It shows that what we’ve been doing previously is almost like working in the dark, where we don't really have that much data, and we are making decisions based on very limited, infrequent data points. So, I think this is a major advance.
