BLOG: Artificial intelligence in early detection of wet AMD

ByJoshua Mali, MD

It is crucial that we employ strategies to better detect intermediate age-related macular degeneration so we can begin treatment sooner and preserve functional vision to, ultimately, help our patients maintain their quality of life.

A recent study examined the prevalence of undiagnosed AMD in a primary eye care setting;

644 patients aged 60 years and older with normal macular health according to their medical record were enrolled. The presence and types of AMD-associated lesions were noted. The investigators found that about 25% of eyes originally thought not to have AMD, based on dilated eye examination by a primary eye care physician, in fact did have macular characteristics that indicated AMD on fundus photography read by trained evaluators.

Baseline visual acuity in AMD predicts visual acuity at 1 and 2 years; individuals who have poor vision at presentation never catch up to their counterparts. Clinical trials and real-world evidence confirm that only a small percentage of patients present with functional vision of 20/40 or greater at choroidal neovascularization (CNV) diagnosis. In fact, some suggest that on average 59 days (± 62) pass between initial symptoms and initiation of treatment, resulting in patients having lost 15 to 25 ETDRS letters at wet AMD diagnosis. These shocking statistics represent the need to improve our detection methods and highlight the critical necessity to catch this conversion to exudative AMD at the earliest possible moment.

Amsler grid or artificial intelligence?

We also know that when the Amsler grid is the only at-home monitoring tool, it has a limited ability to detect visual changes. In addition, there are challenges when it comes to the patients’ ability to accurately utilize the test. It has been shown to have low sensitivity as well as it can be substantially subjective from exam to exam and from patient to patient. According to some reports in the literature, Amsler grid interpretations have poor validity and are unreliable in the clinical diagnosis of retinal defects.

Today, we can do better than relying on what amounts to a lined piece of paper to save our patients’ vision. ForeseeHome (FSH, Notal Vision) uses patented technology that incorporates both artificial intelligence (AI) and telemonitoring, thus allowing patients to test their vision from home and for us to monitor their vision remotely. Patients use the test daily, and their responses are wirelessly transmitted to a data monitoring center and are automatically analyzed by an algorithm that compares them to a normative database and the individual patients’ baseline. Alerts are sent to the physician’s office when significant changes from baseline are detected.

PAGE BREAK

FSH was FDA cleared for home monitoring in 2009. The AREDS2-HOME study compared visual acuity at the time of CNV diagnosis between 1,520 at-risk dry AMD patients who were randomized to the device plus standard of care (self-monitoring with Amsler grid and routine clinic visits) with a control group assigned to standard of care alone. Results demonstrated that among patients who tested with the device at least twice a week, 94% maintained visual acuity of 20/40 greater at diagnosis of wet AMD compared with 62% of eyes that used standard of care alone (P = .014).

The FSH relies on preferential hyperacuity perimetry (PHP). Hyperacuity is our ability as humans to perceive small differences in the relative spatial localization of two objects in space. During the FSH test, stimuli (a series of dots) are successively flashed in various locations along the central 14° of the macula. Most of the dots are aligned with each other and a few dots are intentionally misaligned, creating the perception of a wave or artificial distortion in an otherwise straight line.

If the patient has pathology, the retinal change (CNV) can cause a “competing” distortion. In this scenario, the patient will most likely perceive, and preferentially click, only the location of the greatest distortion. The algorithm recognizes if a patient clicks on the pathological distortion, isolates that area and then quantifies the distortion’s magnitude and location, comparing it to the patient’s baseline.

The FSH is designed to detect morphological changes in the macula that distort vision, including CNV. Other pathologies, however, such as vitreomacular traction, epiretinal membranes and drusen changes/remodeling can also distort vision and cause metamorphopsia. Therefore, not all alerts will lead to a CNV diagnosis. Suspected pathology must be confirmed promptly on a follow-up clinical examination with dilated eye exam, OCT and fluorescein angiogram. I look at a non-CNV alert as another opportunity for me to engage with my patients and reinforce their care. The rate of alerts in the real world is similar to what was observed in the HOME study, or about one alert per patient every 3.4 years.

Artificial intelligence and telemedicine

AI and telemedicine are poised to usher in massive changes in how health care is delivered. AI’s use in telemedicine can be seen in patient monitoring and intelligent assistance diagnosis as well as health care information technology, information analysis and collaboration, and simulation and training systems. AI and telemedicine are symbiotic, and together they can enable us to find diseases earlier, easier and cost-effectively.

By combining remote monitoring with machine learning, our standard of care will reach new and exciting heights and possibly even curtail health care costs. As a retinal specialist, it is truly an honor and privilege to be a leader in this field and be on the forefront of cutting-edge technologies to provide the best possible care for our patients.

References:

AREDS2-HOME Study Research Group, et al. Ophthalmology. 2014;doi:10.1016/j.ophtha.2013.10.027.

Isaac DL, et al. Arq Bras Oftalmol. 2007;doi:10.1590/S0004-27492007000500009.

Midena E, et al. Ophthalmic Res. 2015;doi:10.1159/000441033.

Neely DC, et al. JAMA Ophthalmol. 2017;doi:10.1001/jamaophthalmol.2017.0830.

Rauch R, et al. Retina. 2012;doi:10.1097/IAE.0b013e3182018df6.

Schuchard RA. Arch Ophthalmol. 1993;doi:10.1001/archopht.1993.01090060064024.

Wong TY, et al. Ophthalmology. 2008;doi:10.1016/j.ophtha.2007.03.008.

Disclosure: Mali reports he is a consultant, speaker and stock shareholder for Alimera Sciences, a consultant for and recipient of research funding from Allergan, a consultant and speaker for Genentech, a consultant, speaker and stock shareholder for and recipient of research funding from Regeneron, a consultant and speaker for and recipient of research funding from Notal Vision, a consultant and speaker for Sun Pharmaceutical Industries, and a consultant and speaker for Macular Degeneration Association.