Home OCT with AI analysis may lead to ‘paradigm shift’ in AMD management
Click Here to Manage Email Alerts
A novel artificial intelligence-based algorithm for automatic retinal fluid quantification in home OCT images performed comparably to human readers, according to a presentation from the virtual American Society of Retina Specialists meeting.
This system was able to analyze fluid status, its dynamics and visualize the locations of fluid, Judy E. Kim, MD, professor of ophthalmology and visual sciences at Medical College of Wisconsin, said during the presentation.
“Currently, much of our diagnosis and management of eyes with exudative AMD is dependent on OCT. At-home diagnostic testing systems, such as home OCT, offer an opportunity for personalizing AMD management by providing inter-visit information and allows us to catch the recurrence or progression as soon as it happens,” she said. “This means, we may be able to reduce the likelihood of undertreatment and treatment burden while maintaining visual acuity over the course of treatment.”
Researchers examined the performance of the Notal OCT Analyzer, an AI-based algorithm, in quantifying intraretinal fluid and subretinal fluid and following these volume quantities over time in Notal Home OCT images.
Kim and colleagues collected data from patients with AMD who self-imaged the central 10° of their maculae with the NOCT V2.5 and selected an average of 10 B-scans for manual segmentation. They labeled each B-scan pixel-wise into 4 compartments: vitreous/outer layers (V/O), retina (R), intraretinal fluid (IRF) and subretinal fluid (SRF). Then, they grouped eyes into learning and validation sets (ratio 8:1).
Using semantic segmentation with convolutional neural network, the researchers developed quantifier and assessed the fluid quantification by relating each B-scan’s fluid area segmented by human vs. machine. They also compared pixel-wise fluid with recall/precision, and presence of fluid in B-scans for accuracy, according to the abstract.
The study comprised 355 eyes from 239 participants and 3428 B-scans were manually segmented (75% had fluid). Also, the learning set for the quantifier algorithm development included 2,936 B-scans of 311 eyes and the validation set for the performance evaluation included 492 B-scans of 44 eyes.
Kim reported that the Notal OCT Analyzer performed well compared with the human grader for B-scan segmented fluid detection area (for SRF: Pearson correlation = 0.98 [P < .00001]; for IRF: 0.90 [P < .00001]). The presentation showed that the SRF fluid pixel-wise recall was 0.72 and precision was 0.86, while the IRF recall was 0.8 and the precision was 0.77. The AI-based algorithm also demonstrated high sensitivity and specificity for detecting the presence of SRF (0.99 and 0.98) and IRF (0.99 and 0.97) in B-scans.
“This AI-based analysis of OCT images provides new insights into temporal and retinal fluid dynamics. In addition, we can gain new insights into the markers of disease activity such as the patterns of fluid distribution between intra- and sub-retinal fluid and the changes in fluid volume over time,” Kim said. “We believe the system has the potential to monitor AMD disease activity by the patient at home, which will be a paradigm shift in management of our patients.”
Collapse
Kim JE, et al. Performance of a Novel Deep Learning Algorithm for Automatic Retinal Fluid Quantification in Home OCT Images. Presented at: American Society of Retina Specialists annual meeting; July 24-26, 2020 (virtual meeting).
Click Here to Manage Email Alerts