Purpose: To investigate the correlation of volumetric measurements of intraretinal (IRF) and subretinal fluid obtained by deep learning and central retinal subfield thickness (CSFT) based on optical coherence tomography in retinal vein occlusion, diabetic macular edema, and neovascular age-related macular degeneration.Methods: A previously validated deep learning-based approach was used for automated segmentation of IRF and subretinal fluid in spectral domain optical coherence tomography images. Optical coherence tomography volumes of 2.433 patients obtained from multicenter studies were analyzed. Fluid volumes were measured at baseline and under antivascular endothelial growth factor therapy in the central 1, 3, and 6 mm.Results: Patients with neovascular age-related macular degeneration generally demonstrated the weakest association between CSFT and fluid volume measurements in the central 1 mm (0.107-0.569). In patients with diabetic macular edema, IRF correlated moderately with CSFT (0.668-0.797). In patients with retinal vein occlusion, IRF volumes showed a moderate correlation with CSFT (0.603-0.704). Conclusion:The correlation of CSFT and fluid volumes depends on the underlying pathology. Although the amount of central IRF seems to partly drive CSFT in diabetic macular edema and retinal vein occlusion, it has only a limited impact on patients with neovascular agerelated macular degeneration. Our findings do not support the use of CSFT as a primary or secondary outcome measure for the quantification of exudative activity or treatment guidance.RETINA 42:831-841, 2022O ptical coherence tomography (OCT) has led to profound paradigm shifts in our understanding of retinal disease. The evaluation of leakage and resulting retinal fluid in patients with exudative macular disease using OCT has become a routine task for retina specialists and ophthalmologists around the world. Unfortunately, owing to the large number of patients imaged every day and therefore the amount and detail of image data collected, the feasibility of quantitative manual OCT assessment in busy clinics is inherently impractical and unrealistic. Furthermore, even in research settings, mediocre interobserver reliability and reproducibility have been described as a major dilemma. 1 In effect, physicians merely rely on qualitative biomarkers such as the presence of intraretinal and subretinal fluid when making decisions.Central retinal subfield thickness (CSFT) has traditionally been used in large-scale randomized clinical trials as an indicator for the necessity of antivascular endothelial growth factor (VEGF) treatment in diseases such as retinal vein occlusion (RVO), diabetic

No abstract

Functional magnetic resonance imaging (fMRI) combined with population receptive field (pRF) mapping allows for associating positions on the visual cortex to areas on the visual field. Apart from applications in healthy subjects, this method can also be used to examine dysfunctions in patients suffering from partial visual field losses. While such objective measurement of visual deficits (scotoma) is of great importance for, e.g., longitudinal studies addressing treatment effects, it requires a thorough assessment of accuracy and reproducibility of the results obtained. In this study, we quantified the reproducibility of pRF mapping results within and across sessions in case of central visual field loss in a group of 15 human subjects. We simulated scotoma by masking a central area of 2° radius from stimulation to establish ground-truth conditions. This study was performed on a 7T ultra-high field MRI scanner for increased sensitivity. We found excellent intrasession and intersession reproducibility for the pRF center position (Spearman correlation coefficients for x , y : >0.95; eccentricity: >0.87; polar angle: >0.98), but only modest reproducibility for pRF size (Spearman correlation coefficients around 0.4). We further examined the scotoma detection performance using an automated method based on a reference dataset acquired with full-field stimulation. For the 2° artificial scotoma, the group-averaged scotoma sizes were estimated at between 1.92° and 2.19° for different sessions. We conclude that pRF mapping of visual field losses yields robust, reproducible measures of retinal function and suggest the use of pRF mapping as an objective method for monitoring visual deficits during therapeutic interventions or disease progression.

Purpose To evaluate the reliability of automated fluid detection in identifying retinal fluid activity in OCT scans of patients treated with anti-VEGF therapy for neovascular age-related macular degeneration by correlating human expert and automated measurements with central retinal subfield thickness (CSFT) and fluid volume values. Methods We utilized an automated deep learning approach to quantify macular fluid in SD-OCT volumes (Cirrus, Spectralis, Topcon) from patients of HAWK and HARRIER Studies. Three-dimensional volumes for IRF and SRF were measured at baseline and under therapy in the central millimeter and compared to fluid gradings, CSFT and foveal centerpoint thickness (CPT) values measured by the Vienna Reading Center. Results 41.906 SD-OCT volume scans were included into the analysis. Concordance between human expert grading and automated algorithm performance reached AUC values of 0.93/0.85 for IRF and 0.87 for SRF in HARRIER/HAWK in the central millimeter. IRF volumes showed a moderate correlation with CSFT at baseline (HAWK: r = 0.54; HARRIER: r = 0.62) and weaker correlation under therapy (HAWK: r = 0.44; HARRIER: r = 0.34). SRF and CSFT correlations were low at baseline (HAWK: r = 0.29; HARRIER: r = 0.22) and under therapy (HAWK: r = 0.38; HARRIER: r = 0.45). The residual standard error (IRF: 75.90 µm; SRF: 95.26 µm) and marginal residual standard deviations (IRF: 46.35 µm; SRF: 44.19 µm) of fluid volume were high compared to the range of CSFT values. Conclusion Deep learning-based segmentation of retinal fluid performs reliably on OCT images. CSFT values are weak indicators for fluid activity in nAMD. Automated quantification of fluid types, highlight the potential of deep learning-based approaches to objectively monitor anti-VEGF therapy.

No abstract