A smartphone that we use to view a fundus image.
Reviewed by Divya Parthasarathy Rao, MD
Glaucoma is a major sight-threatening problem worldwide, and in developing countries the threat to eye health is even greater because the disease goes undetected in large numbers of patients.
According to Divya Parthasarathy Rao, MD, of Remidio Innovative Solutions Inc, USA, more than 90% of cases go undetected. After detection, 50% already have advanced disease and 20% are already blind.
Glaucoma screening has taken a big step forward in solving this challenge with the development of a new automated artificial intelligence (AI) tool, AI-Glaucoma Media. Performance is promising and it is offline capable, meaning it can be used at the edge without the need for internet access or a cloud-based interface, with reports in less than 10 seconds, according to Rao.
Prospectively validated automated technology, he noted, may expand the reach of glaucoma care to underserved regions.
Testing the AI device
Rao and colleagues evaluated the diagnostic capabilities of a smartphone-based fundus camera with an offline artificial intelligence tool for glaucoma detection in a prospective study of 243 subjects at a tertiary glaucoma center in the India. The study was conducted between July 2021 and February 2022. All participants had glaucoma of varying severity, were glaucoma suspects, or served as normal controls.
The researchers captured disc-centric images using a clinically well-validated portable fundus camera. The aim of the study was to determine if this tool can detect referable glaucoma, defined as defined structural and functional retinal changes related to glaucoma. The results obtained with the device were compared with the final diagnoses made by specialists after a glaucoma study that included a clinical evaluation, a spectral domain optical coherence tomography and a Humphrey visual field evaluation.
The tool performed well compared to the results of a full glaucoma study conducted by a glaucoma specialist.
“Among the 243 patients, the sensitivity of the AI tool for detecting remittable glaucoma was 93.7% (95% confidence interval [CI], 87.6%-96.9%), the specificity was 85.6% (95% CI, 78.6%-90.6%), the positive predictive value was 84.6%, and the value negative predictive was 94.2%,” the researchers wrote. They also noted that there was minimal overreference of normal subjects (‘no glaucoma’ recall was 94.7%; 95% CI, 87.2–97.9%).
Compared to image classification performed by just 3 glaucoma specialists, the results were impressive. The specialists detected 60% of patients with true glaucoma versus 94% detected by the AI tool using the same images.
“AI may have learned to detect subtle structural changes that are not apparent to the human eye,” Rao said.
Following this study, the company will test the AI tool in a prospective, real-world study in a community, with the goal of testing its performance based on glaucoma severity. The study under discussion was limited to a population in South Asia, but the next step in the evaluation will include patients from multiple ethnic groups.
Based on the current results, the researchers consider the AI tool’s ability to detect glaucoma “promising.”
A key advantage of this device is that it can be incorporated into an inexpensive, portable camera without compromising efficiency and accuracy. It also provides instant results and may have the potential to have a major impact in the detection of this vision-threatening disease. Additionally, it can be used by healthcare workers in communities with limited resources to provide much-needed eye care.
Dr Divya Parthasarathy Rao
Rao is an employee of Remidio Innovative Solutions Inc, USA.