Smartphone AI prediction tool

A novel portable smartphone slit-lamp prototype system employing artificial intelligence (AI) is showing promise as a low-cost screening tool for angle-closure disease and could be suitable for use in the primary care setting, said David Chen FRCOphth, Department of Ophthalmology, National University Hospital, Singapore. “The portable smartphone slit lamp prototype provides strong positive correlation with selected anterior chamber measurements taken with anterior segment OCT,” Dr Chen told the 37th Congress of the ESCRS in Paris, France. The new device called the Mobile Imaging Device for Anterior Segment (MIDAS) imaging consists of a portable slit-lamp prototype that can be attached to most of the major brands of smartphones. It is designed to assess anterior chamber depth of phakic patients with an undilated pupil. The system uses artificial intelligence to show correlations between the anterior chamber parameters predicted by MIDAS and measurements made with anterior segment optical coherence tomography (AS-OCT). In a prospective, single-centre image-validation study, Dr Chen and associates performed sequential image capture using two devices, the MIDAS with Samsung Galaxy S7 smartphone and the Tomey SS-1000 CASIA (AS-OCT) device. They scanned 49 eyes of 49 patients, three of whom had clinically detectable angle-closure disease. The patients were older than 60 with no history of laser or intraocular surgery. The study showed that the three angle-closure eyes in the study had significantly different anterior chamber parameters than the remaining 46 without angle closure. The mean temporal angle opening distance (AOD-500) was 91.5µm in narrow angle eyes, compared to 286.0µm in normal eyes (p<0.01). The mean central anterior chamber depth (ACD) was 1746µm in narrow angle eyes, compared to 2694µm in normal eyes (p<0.01). All three angle-closure eyes were successfully predicted by MIDAS. There was also a strong correlation between the ACD predicted by MIDAS and those measured by the CASIA AS-OCT. A Bland-Altman plot showed more than 50% of predicted ACDs were within 20µm (1%) of measured ACDs and more than 95% of predicted ACDs were within 200µm (10%) of measured ACDs. “In this proof of concept study, this is the first AI developed on a portable slit lamp device that successfully predicted central anterior chamber depth measurements of asymptomatic patients. Recruitment for more patients and data points is under way,” he concluded. David Chen: dzychen18@gmail.com

Tags: artificial intelligence glaucoma