Dataset Information

Accuracy of a Machine-Learning Algorithm for Detecting and Classifying Choroidal Neovascularization on Spectral-Domain Optical Coherence Tomography.

ABSTRACT:

Background

To evaluate the performance of a machine-learning (ML) algorithm to detect and classify choroidal neovascularization (CNV), secondary to age-related macular degeneration (AMD) on spectral-domain optical coherence tomography (SD-OCT) images.

Methods

Baseline fluorescein angiography (FA) and SD-OCT images from 1037 treatment-naive study eyes and 531 fellow eyes, without advanced AMD from the phase 3 HARBOR trial (NCT00891735), were used to develop, train, and cross-validate an ML pipeline combining deep-learning-based segmentation of SD-OCT B-scans and CNV classification, based on features derived from the segmentations, in a five-fold setting. FA classification of the CNV phenotypes from HARBOR was used for generating the ground truth for model development. SD-OCT scans from the phase 2 AVENUE trial (NCT02484690) were used to externally validate the ML model.

Results

The ML algorithm discriminated CNV absence from CNV presence, with a very high accuracy (area under the receiver operating characteristic [AUROC] = 0.99), and classified occult versus predominantly classic CNV types, per FA assessment, with a high accuracy (AUROC = 0.91) on HARBOR SD-OCT images. Minimally classic CNV was discriminated with significantly lower performance. Occult and predominantly classic CNV types could be discriminated with AUROC = 0.88 on baseline SD-OCT images of 165 study eyes, with CNV from AVENUE.

Conclusions

Our ML model was able to detect CNV presence and CNV subtypes on SD-OCT images with high accuracy in patients with neovascular AMD.

SUBMITTER: Maunz A

PROVIDER: S-EPMC8227725 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:PURPOSE:To detect and quantify choroidal neovascularization (CNV) in patients with age-related macular degeneration (AMD) using optical coherence tomography (OCT) angiography. DESIGN:Observational, cross-sectional study. PARTICIPANTS:A total of 5 normal subjects and 5 subjects with neovascular AMD were included. METHODS:A total of 5 eyes with neovascular AMD and 5 normal age-matched controls were scanned by a high-speed (100 000 A-scans/seconds) 1050-nm wavelength swept-source OCT. The macular angiography scan covered a 3 × 3-mm area and comprised 200 × 200 × 8 A-scans acquired in 3.5 seconds. Flow was detected using the split-spectrum amplitude-decorrelation angiography (SSADA) algorithm. Motion artifacts were removed by 3-dimensional (3D) orthogonal registration and merging of 4 scans. The 3D angiography was segmented into 3 layers: inner retina (to show retinal vasculature), outer retina (to identify CNV), and choroid. En face maximum projection was used to obtain 2-dimensional angiograms from the 3 layers. The CNV area and flow index were computed from the en face OCT angiogram of the outer retinal layer. Flow (decorrelation) and structural data were combined in composite color angiograms for both en face and cross-sectional views. MAIN OUTCOME MEASURES:The CNV angiogram, CNV area, and CNV flow index. RESULTS:En face OCT angiograms of CNV showed sizes and locations that were confirmed by fluorescein angiography (FA). Optical coherence tomography angiography provided more distinct vascular network patterns that were less obscured by subretinal hemorrhage. The en face angiograms also showed areas of reduced choroidal flow adjacent to the CNV in all cases and significantly reduced retinal flow in 1 case. Cross-sectional angiograms were used to visualize CNV location relative to the retinal pigment epithelium and Bruch's layer and classify type I and type II CNV. A feeder vessel could be identified in 1 case. Higher flow indexes were associated with larger CNV and type II CNV. CONCLUSIONS:Optical coherence tomography angiography provides depth-resolved information and detailed images of CNV in neovascular AMD. Quantitative information regarding CNV flow and area can be obtained. Further studies are needed to assess the role of quantitative OCT angiography in the evaluation and treatment of neovascular AMD.

Project description:This study aimed to investigate the differences between images obtained by optical coherence tomography angiography (OCTA) with those from immunohistochemical labeling of laser-induced choroidal neovascularization (CNV) in a mouse model. CNV was induced by laser photocoagulation (GYC-2000, NIDEK; wavelength 532 nm) in the left eyes of 10 female C57BL/6J mice aged 6 weeks. The laser parameters included a 100-μm spot, 100-ms pulse duration and 200-mW incident power to rupture Bruch's membrane. OCT and OCTA CNV images were obtained using the RS-3000 Advance (NIDEK) 5 days post-laser photocoagulation. After OCTA imaging, the isolated choroid/retinal pigment epithelium complexes were fluorescently labeled with CD31 (an endothelial cell marker), platelet-derived growth factor receptor β (PDGFRβ, a pericyte-like scaffold marker), α-smooth muscle actin (α-SMA) and collagen I. Area measurements of the lesions obtained by enface OCTA were compared with immunolabeled CD31+ CNV lesions in choroid flat-mounts. We also examined structural correlations between the PDGFRβ+ pericyte-like scaffold and OCTA images. Laser-induced CNV was clearly detected by enface OCTA, appearing as a hyperflow lesion surrounded by a dark halo. Area measurements of the CNV lesion by immunolabeling were significantly larger than those obtained by enface OCTA (p = 0.006). The CNV lesion beneath the periphery of the pericyte-like scaffold was not clearly visible by enface OCTA due to the dark halo; however, the lesion was detectable as blood flow by cross-sectional OCTA and was also highly labeled by CD31. The periphery of the pericyte-like scaffold appeared to develop into subretinal fibrosis and this region was rich in myofibroblasts. Enface OCTA was unable to detect the entire area of laser-induced CNV in mice, with an undetectable portion located beneath the fibrotic periphery of the pericyte-like scaffold. Due to this OCTA fibrosis artifact, OCTA imaging has limited potential for accurately estimating CNV lesions.