Longitudinal Validation of the Artificial Intelligence Algorithm in Home OCT for Age-Related Macular Degeneration—Report 3

• Published in: Ophthalmology science (Online) p. 100907
• Main Authors: Leng, Theodore, Leung, Ella H., Mukkamala, S. Krishna, Taban, Mehran R., Havilio, Moshe, Nahen, Kester, Mohan, Nishant, Benyamini, Gidi, Keenan, Tiarnan D.L.
• Format: Journal Article
• Language: English
• Published: Elsevier Inc 01.08.2025
• Subjects: Artificial intelligence; Home OCT; Neovascular age-related macular degeneration; OCT; Ophthalmology; Optical coherence tomography; Reference change value; OCT; ERM; NOA; RCV; Reference change value; OT; ROC; AI; HRS; AUROC; SD; BCVA; SNR; Neovascular age-related macular degeneration; Home OCT; nAMD; HOCT; Artificial intelligence; Total retinal hypo-reflective spaces; Anti-vascular endothelial growth factor; Optical coherence tomography; Volume units; Notal Health Cloud; artificial intelligence; Notal OCT Analyzer; TRO; home OCT; In-office OCT; Signal to Noise Ratio; Nanoliter; AMD; Hypo-reflective spaces; IO-OCT; nL; neovascular age-related macular degeneration; Neovascular AMD; VEGF; Best-corrected visual acuity; Spectral-domain OCT; SD-OCT; Age-related macular degeneration; Cloud; reference change value; VU
• ISSN: 2666-9145; 2666-9145
• DOI: 10.1016/j.xops.2025.100907

Longitudinal validation of the artificial intelligence–based Notal OCT Analyzer (NOA) for identification of clinically significant changes in the trajectories of retinal total hyporeflective spaces (TRO) from daily home OCT (HOCT) self-imaging in eyes with age-related macular degeneration. Post hoc analysis of the HOCT Fluid Visualization Agreement Study. Three hundred seventeen eyes of 180 participants who self-imaged daily using the HOCT for 5 weeks. For each eye study, the ground truth of TRO stability or change was defined by human experts grading the 5-week time series of HOCT volume scans. The TRO trajectory of the 5 weeks was plotted separately for each study eye by NOA. Three approaches to identifying the optimal threshold (OT) for a clinically significant change in TRO were pursued: (1) personalized approach based on the reference change value methodology used in laboratory medicine; (2) optimized uniform approach; (3) uniform approach commonly used of 10 volume units (VU). The personalized approach comprised TRO curve-fitting to evaluate the change in amplitude (signal) and within-subject variations (noise), followed by receiver operating characteristic analysis of the signal-to-noise ratio (SNR) to identify the OT for determination of a clinically significant change in TRO. Area under the ROC curve (AUROC); sensitivity, specificity, and accuracy at the OT. Of the 296 trajectories analyzed, 107 (36.1%) were classified as changing and the remaining 189 (63.9%) as stable. The personalized approach had AUROC of 0.9811 (with 99.1% sensitivity, 89.4% specificity, and 94.2% accuracy), at OT of SNR = 2.42. The optimized uniform approach had AUROC of 0.9687 (with 94.4% sensitivity, 89.4% specificity, and 91.9% accuracy), at OT of 3.88 VU. The 10 VU uniform approach had 76.6% sensitivity, 94.7% specificity, and 85.7% accuracy. The NOA-generated trajectories permitted highly accurate determination of TRO change vs. stability, with favorable SNR. The personalized approach had higher sensitivity in detecting TRO change at 99.1% than the uniform approaches. Notal OCT Analyzer and its trajectories were a reliable tool for identifying clinically relevant changes in retinal fluid status by the quality standards of laboratory medicine. Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.