Related Experiment Video
Updated: May 19, 2026
Evaluation of Capillary and Other Vessel Contribution to Macular Perfusion Density Measured with Optical Coherence Tomography Angiography
Published on: February 18, 2022
Clinical application of automated optical coherence tomography angiography for retinal fluid segmentation: a study of
Anhai Wei1,2, Yangchen Zhou3, Rui Nie1
1Department of Medical Engineering, Daping Hospital, Army Medical University, Chongqing, China.
Background:
Precise quantification of subretinal fluid (SRF) is crucial for the diagnosis, management, and prognosis of central serous chorioretinopathy (CSC), yet the automated SRF segmentation function of the Intalight VG100 optical coherence tomography angiography (OCTA) device lacks real-world clinical validation. This study aimed to clinically validate the Intalight VG100 OCTA device's automated segmentation of SRF area and volume against physician annotations.
Methods:
A retrospective analysis was conducted on OCTA images from 78 patients with CSC (139 eyes). The device's automated segmentation SRF area and volume were quantified via the VG100's companion software VanGoghReview, and physician-annotated segmentation served as the reference gold standard. The Wilcoxon signed-rank test, Spearman rank correlation analysis, and Bland-Altman analysis were used to compare quantitative differences, assess correlations, and evaluate measurement agreement, respectively. Additionally, SRF area and volume error distributions were analyzed, with assessments stratified by lesion size (area: 0 mm2 ≤ area <5 mm2, 5 mm2 ≤ area <10 mm2, and area ≥10 mm2; volume: 0 mm3 ≤ volume <0.5 mm3, 0.5 mm3 ≤ volume <1 mm3, and volume ≥1 mm3).
Results:
No significant difference in SRF area was noted between device's automated segmentation and the physician-annotated segmentation (P>0.05), while volume was significantly overestimated by the device (P<0.05). Both parameters exhibited excellent correlations (area: Spearman ρ=0.981; volume: ρ=0.995; both P values <0.05). Bland-Altman analysis revealed a low proportion of outliers, with 6.47% of samples for SRF area and 1.44% for SRF volume falling outside the 95% limits of agreement (LoA). Compared to physician annotations, the device's automated segmentation underestimated SRF area by a mean of 0.25 mm2 while overestimating SRF volume by 0.11 mm3. In contrast, the performance for area segmentation was significantly better: 88% of absolute errors were within 0-2 mm2, and 81% of relative errors were <20%. For volume, 91% of absolute errors were within 0-0.2 mm3, but only 40% of relative errors were <20%. Small-to-moderate lesions (area <10 mm2) showed high consistency in area segmentation, while large lesions (area ≥10 mm2) exhibited more dispersed errors. Volume segmentation errors were scattered regardless of lesion size, with error dispersion increasing with larger lesion size.
Conclusions:
This study is the first to evaluate the clinical application of the Intalight VG100's automated segmentation SRF with real-world data. Its SRF area quantification is reliable for small-to-moderate lesions, supporting routine clinical monitoring of CSC. In contrast, the device consistently overestimates volume quantification and yields considerable variability, necessitating manual verification for large lesions. These findings confirm the device's clinical utility and identify key directions for algorithm optimization.
