PINNACLE trial uses OCT‑AI phenotyping

Retinal OCT scans are already the workhorse for age-related macular degeneration. But intermediate AMD is the frustrating middle zone — patients clearly have disease, yet clinicians still struggle to say which eyes are quietly stable and which ones are lining up for geographic atrophy or neovascular AMD. (bjo.bmj.com) The new PINNACLE paper, published online May 7 in the *British Journal of Ophthalmology*, pushes on that gap by using expert grading plus AI-based OCT analysis to map the baseline anatomy of one of the field’s biggest prospective intermediate-AMD cohorts. (bjo.bmj.com) ### What is PINNACLE trying to solve? PINNACLE is a prospective, multicenter study built to predict which patients with intermediate AMD will progress to late disease. (nature.com) The broader project combines a huge retrospective imaging set — more than 400,000 OCT images — with prospective follow-up every 4 months in patients aged 55 to 90 across sites in the UK, Austria, and Switzerland. The whole point is to turn vague “higher risk” language into something more individualized. ### Why use OCT and AI together? OCT shows the retina in cross-section, layer by layer. That is where many of AMD’s most useful warning signs live. But human grading gets noisy fast when you try to measure lots of subtle features across large cohorts. (bjo.bmj.com) The new report mixes expert reads with AI-based image analysis so the team can quantify prevalence, amount, location, and overlap of features instead of just noting whether they exist. ### What features did they track? The paper focused on the usual suspects in intermediate AMD, but in a more structured way — drusen and drusen subtypes, subretinal drusenoid deposits, hyperreflective foci, double-layer sign, and markers of outer retinal damage like ellipsoid-zone and outer nuclear layer thickness. (nature.com) It also included incomplete RPE and outer retinal atrophy and a semi-automated atrophic marker built from ellipsoid-zone loss plus choroidal hypertransmission. Basically, this is a much denser anatomical inventory than what most clinic notes capture. ### What did the new report actually show? The headline result is that these lesions are not randomly scattered. (bjo.bmj.com) Drusen concentrated within the central 3 mm of the macula, while subretinal drusenoid deposits showed up mainly in the superior perifoveal quadrant. Higher drusen volume traveled with the presence and volume of hyperreflective foci, but larger drusen burden was inversely correlated with subretinal drusenoid deposits. That starts to look less like one disease marching in lockstep and more like several structural patterns sharing the same diagnostic label. ### Why do the outer retinal measurements matter? Because they get closer to function. The paper showed thinning in the ellipsoid zone and outer nuclear layer — signs of photoreceptor compromise — and that thinning was worse in eyes already showing atrophic features such as iRORA or the ellipsoid-zone-loss/hypertransmission marker. (bjo.bmj.com) In plain English, once those structural damage flags appear, the retina is not just decorated differently — it looks more biologically stressed. ### Is this already a risk score? Not yet. This report is mainly a baseline map, not the final prognostic engine. But it matters because PINNACLE was designed from the start to validate imaging biomarkers that predict progression, and this kind of standardized phenotyping is the groundwork for that. (bjo.bmj.com) If you cannot define the patterns cleanly, you cannot build a reliable progression model on top of them. ### Why should clinicians care now? Because intermediate AMD follow-up is still pretty blunt. Many clinics monitor on broad intervals and escalate when obvious change appears. A cleaner OCT phenotype could eventually help separate eyes that need tighter surveillance from eyes that can be watched less aggressively, and it could also make trial enrollment smarter by enriching for patients whose anatomy already signals higher risk. (bjo.bmj.com) The paper itself frames the dataset as useful for interpreting longitudinal data and refining future trial inclusion criteria. ### Bottom line This is not the moment AI “solved” AMD triage. But it is a real step toward turning intermediate AMD from a mushy category into a set of measurable retinal patterns — and that is the prerequisite for better prediction, better trials, and eventually better follow-up decisions. (nature.com) (bjo.bmj.com)