Utilization of deep learning to quantify fluid volume of neovascular age-related macular degeneration patients based on swept-source OCT imaging: The ONTARIO study

Abstract

Purpose To evaluate the predictive ability of a deep learning-based algorithm to determine long-term best-corrected distance visual acuity (BCVA) outcomes in neovascular age-related macular degeneration (nARMD) patients using baseline swept-source optical coherence tomography (SS-OCT) and OCT-angiography (OCT-A) data. Methods In this phase IV, retrospective, proof of concept, single center study, SS-OCT data from 17 previously treated nARMD eyes was used to assess retinal layer thicknesses, as well as quantify intraretinal fluid (IRF), subretinal fluid (SRF), and serous pigment epithelium detachments (PEDs) using a novel deep learning-based, macular fluid segmentation algorithm. Baseline OCT and OCT-A morphological features and fluid measurements were correlated using the Pearson correlation coefficient (PCC) to changes in BCVA from baseline to week 52. Results Total retinal fluid (IRF, SRF and PED) volume at baseline had the strongest correlation to improvement in BCVA at month 12 (PCC = 0.652, p = 0.005). Fluid was subsequently sub-categorized into IRF, SRF and PED, with PED volume having the next highest correlation (PCC = 0.648, p = 0.005) to BCVA improvement. Average total retinal thickness in isolation demonstrated poor correlation (PCC = 0.334, p = 0.189). When two features, mean choroidal neovascular membranes (CNVM) size and total fluid volume, were combined and correlated with visual outcomes, the highest correlation increased to PCC = 0.695 (p = 0.002). Conclusions In isolation, total fluid volume most closely correlates with change in BCVA values between baseline and week 52. In combination with complimentary information from OCT-A, an improvement in the linear correlation score was observed. Average total retinal thickness provided a lower correlation, and thus provides a lower predictive outcome than alternative metrics assessed. Clinically, a machine-learning approach to analyzing fluid metrics in combination with lesion size may provide an advantage in personalizing therapy and predicting BCVA outcomes at week 52.

The Architecture of Macular Neovascularizations Predicts Treatment Responses to Anti-VEGF Therapy in Neovascular AMD

“…MNV with treatment-resistant SRF seem to exert a protective effect on the growth of geographic atrophy and are associated with better vision, so under certain circumstances, SRF is tolerated [ 22 , 36 ]. PED, however, are associated with poorer vision in the long term, especially if the RPE tears in the course of the disease [ 37 ]. The presence of IRF also goes along with a negative effect on visual outcome [ 22 , 28 ].…”

Quantitative Assessment of Automated Optical Coherence Tomography Image Analysis Using a Home-Based Device for Self-Monitoring Neovascular Age-Related Macular Degeneration

“…This U-Net like architecture 13 - an autoencoder with skip connections - takes as input both the OCT B-scans and a retinal layer segmentation mask from Orion and has been successfully applied to similar image segmentation tasks. 14…”

“…This U-Net like architecture 13 -an autoencoder with skip connections -takes as input both the OCT B-scans and a retinal layer segmentation mask from Orion and has been successfully applied to similar image segmentation tasks. 14 Learning is based on minimizing the model's loss, where the loss function weights categorical cross entropy with the DSC. A 10-fold cross-validation was used to evaluate all scans from all patient eyes, where, importantly, folds were stratified by the patient.…”

Optical Coherence Tomography Angiography: A 2023 Focused Update on Age-Related Macular Degeneration

“…Errors can also be associated with low contrast between neovascularization vessels and the background signal, requiring time-consuming review of images [ 96 ]. Recently, DL systems incorporated OCTA into the model training for computer-aided diagnosis (CAD) of retinal diseases, including AMD [ 96 – 100 ]. Recently, Wang and colleagues proposed a CNN-based model by using OCTA real-world multicenter dataset for MNV diagnosis and segmentation in AMD and non-AMD pathologies.…”