Q&A: New studies with DARC technology show potential for early detection of wet AMD
In a paper in Expert Review of Molecular Diagnostics, Maria Francesca Cordeiro, MD, and colleagues demonstrated that algorithm-aided detection of apoptosing retinal cells can identify early endothelial neovascular activity.
This method predicted onset of wet age-related macular degeneration several months in advance with a specificity of 79% to 90% and sensitivity greater than 80%.
In a previous study, the same method showed efficacy in predicting glaucomatous progressive disease. This way of testing the eye could be the answer to the unmet need of diagnosing AMD early, before significant vision deterioration occurs, to get the best results from treatment.
Cordeiro is professor of ophthalmology at Imperial College and University College London and director of ICORG clinical trials unit at Western Eye Hospital. She developed the detection of apoptosing retinal cells (DARC) imaging technology several years ago specifically for glaucoma and over time has continued her research, finding several new applications and more recently combining it with an artificial intelligence approach. In this interview with Healio/OSN, she spoke about DARC and the study that assessed and confirmed its potential as a biomarker for wet AMD.
Question: What is DARC?
Answer: Basically, DARC is a technology that takes advantage of the transparency of the eye. As we use the eye to look out, we can use it also to look within, down to the level of single cells. By means of confocal scanning laser ophthalmoscopy, DARC visualizes apoptotic and stressed retinal cells in vivo. In the case of AMD, the stressed cells are those that are going to form new blood vessels. The technology has evolved over the last 15 years, coming from concept stage to clinical stage. Phase 1 and 2 clinical studies have shown that it is safe and effective. The dye we use is different from the dyes used in fluorescein angiography or even indocyanine green angiography (ICGA). It stems from an endogenous protein, which is fluorescently labeled to allow us to visualize the stressed cells in the back of the eye.
Q: What is the potential role of DARC in AMD?
A: Given the positive results in glaucoma, in which our DARC convolutional neural network (CNN) algorithm was able to predict glaucoma progression 18 months in advance, we have used the same DARC CNN, with no modifications or retraining, to assess wet AMD. We looked specifically at patients who at the baseline visit already had wet or established dry AMD. They had one assessment with DARC initially, and then we were able to follow them up for another 3 years with repeated OCT scans. OCT assessment is something we do routinely with AMD patients, and we were able to match up the DARC signal with serial OCT scans and show that DARC spots, detected 36 months in advance, predicted where new subretinal fluid formed on OCT, indicating new vessel activity and leakage. DARC can detect this with an accuracy that was unheard of before. It can locate labeled DARC cells that go on to show new fluid formed in the subretinal space up to 3 years in advance.
Q: Please elaborate on the study.
A: This phase 2 clinical trial of DARC was conducted at Western Eye Hospital and included 29 eyes of 18 patients with baseline evidence of either wet or dry AMD. In the first 6 months of 2017, patients received a single intravenous injection of fluorescent annexin V and had DARC performed once only at baseline. They were then followed up with OCT every 6 month for 3 years. DARC images of each eye were aligned with the corresponding OCT scans, allowing for the location of a DARC spot to be identified on the corresponding OCT scan at subsequent examinations. At each time point, we assessed whether a DARC spot from baseline was associated with the occurrence of wet AMD. DARC CNN was able to predict the development of new subretinal fluid over a 36-month period with a specificity and sensitivity of more than 70%. This was important to us because we learned from the latest AREDS study that detecting subretinal fluid is not easy for clinicians. When they compared the accuracy of a deep learning algorithm on OCT sequential scans with observation by expert clinicians, the deep learning algorithm performed better. With our CNN algorithm, we were able to look individually at more than 20,000 OCT slices to see if we could detect subretinal fluid. This is something that could not be easily achieved by even the best human experts.
Q: What is the clinical significance of this study?
A: Our findings can affect clinical practice in three ways. First, because there are patients who even on a proactive treat-and-extend regimen are at a high risk of disease recurrence or progression, we need to identify them and treat them more intensively early on so that they do not lose vision. Second, in the fellow eye, DARC can show the early signs of the disease well in advance of conventional methods — even before it manifests in reducing vision. The third way it can affect our practice is by offering a new screening method that not only ophthalmologists but also optometrists and properly trained technicians can use, targeting those older than the age of 70 years. The presence of DARC spots around the macula could be an early warning sign of developing new wet AMD.
Q: What are the future steps in the development and applications of DARC?
A: At the moment, DARC requires intravenous injection of the dye and is therefore limited to medically trained personnel, but we have already developed a nasal-inhaled version, which we hope to roll out soon. In the future, subjects can self-administer just before the fundus photograph is taken by an optometrist or a technician. In addition, the florescence wavelength chosen for this biomarker is exactly the same as for ICGA, so anyone who has the equipment for reading ICGA will be able to use it also for DARC. Our aim is to make DARC widely accessible and noninvasive. At the moment, it is used in trials as an exploratory endpoint for testing new drugs. The approval as a diagnostic test would require a larger multicenter phase 3 trial comparing DARC with OCT scans and other gold standards. At this stage, we are still collecting data, and this is helping us to increase the accuracy of our AI algorithm. In the long term, we also hope to further validate DARC in neurodegenerative conditions including Alzheimer’s and Parkinson’s.
- Corazza P, et al. Expert Rev Mol Diagn. 2020;doi:10.1080/14737159.2020.1865806.
- Normando EM, et al. Expert Rev Mol Diagn. 2020;doi:10.1080/14737159.2020.1758067.