Correlation between human expert macular fluid height assessment and fluid volume quantification in neovascular age-related macular degeneration

Correlation between human expert macular fluid height assessment and fluid volume quantification in neovascular age-related macular degeneration | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Correlation between human expert macular fluid height assessment and fluid volume quantification in neovascular age-related macular degeneration Stefan Steiner, Bianca S. Gerendas, Gabor Deak, Oliver Leingang, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5313889/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 31 Mar, 2026 Read the published version in Scientific Reports → Version 1 posted 11 You are reading this latest preprint version Abstract To investigate the association of manually measured retinal fluid by a human expert with AI-quantified retinal fluid volumes and to explore disease activity indicated by fluid volume distributions in neovascular age-related macular degeneration (nAMD) using an approved AI-based algorithm. This is a retrospective study analyzing baseline OCT data of patients with nAMD from multicenter study data. Manually measured maximum macular fluid heights vertically on B-scans in the central millimeter (CMM) for intraretinal fluid (IRF), subretinal fluid (SRF), and pigment epithelial detachment (PED) acquired from expert were associated with vertical fluid heights and three-dimensional volumes obtained by automated quantification using an AI-based tool (RetInSight Fluid Monitor Version 2). Out of 890 patients/eyes, we identified IRF in the CMM both manually and automatically in 328 eyes, SRF in 502 eyes, and PED in 705 eyes. The correlation between manual height and AI-based height was strong for IRF (r = 0.87) and PED (r = 0.91), and moderate for SRF (r = 0.67). Manual height vs. AI-based volume correlation in the CMM was strong for IRF (r = 0.76), and PED (r = 0.87) and moderate for SRF (r = 0.55). The correlation worsened when associating total fluid volumes in the central 6mm with manual CMM fluid height, indicating that CMM height does not represent total nAMD disease activity. AI-based fluid segmentation, in contrast to conventional human expert fluid measurements, provides a more comprehensive assessment, allowing for a significantly more accurate interpretation of total nAMD disease activity. Health sciences/Diseases/Eye diseases/Macular degeneration Health sciences/Diseases/Eye diseases/Retinal diseases age-related macular degeneration artificial intelligence deep learning fluid quantification optical coherence tomography intraretinal fluid subretinal fluid pigment epithelial detachment Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction Age-related macular degeneration (AMD) is a widespread and debilitating ocular condition primarily affecting the elderly, with a growing impact not only on patients but also on ophthalmologists, insurance companies, and healthcare providers, due to increased life expectancy.[ 1 – 3 ] Neovascular AMD (nAMD) is characterized by an increased release of pro-angiogenic factors such as vascular endothelial growth factor (VEGF), leading to macular neovascularization (MNV) resulting in the accumulation of fluid in different retinal layers which causes vision impairment and distortion. Gold-standard for treating nAMD is the application of intravitreal anti-VEGF, with the aim to reduce the amount and recurrence of retinal fluid and prevent irreversible scaring of the macula. Accumulation of intraretinal (IRF) and subretinal fluid (SRF) are the main drivers in clinical decision-making whether a patient needs anti-VEGF administration.[ 4 – 6 ] With optical coherence tomography (OCT) as advanced retinal imaging technology macular fluid compartments like IRF, SRF, and pigment epithelium detachment (PED) can be visualized in high resolution. Nevertheless, so far it has been time-consuming and challenging to quantify the amount of three-dimensional retinal fluids present in volumes rather than individual B-scans. Manual reading center measurements (e.g., central retinal thickness, central subfield thickness, and center point thickness) have historically been used as surrogate markers for the accumulation of retinal fluids, disease activity and were the base for dosing regimen in clinical trials.[ 7 – 9 ] These semi-quantitative indicators were introduced since time-domain OCT lacked the raster scanning capability of today’s OCT technology not able to perform fluid quantifications throughout the entire volume. In addition, available evidence suggests that these semi-quantitative measurements are poor indicators of total disease activity.[ 10 , 11 ] Artificial intelligence (AI) algorithms can precisely identify and segment individual fluid compartments in nAMD from entire OCT volumes in a fraction of time compared to a human counting fluid associated pixels, on high numbers of B-scans, allowing for the detection and quantification of nAMD-related exudation. [ 12 ] Visualization and quantification of AI-based features support ophthalmologists in detecting and monitoring retinal fluid over time, enabling personalized treatment and efficient therapeutic interventions. By automating retinal segmentation, AI reduces the burden on healthcare professionals, enhances diagnostic precision, while improving workflows and potentially enhancing patient care and visual outcomes for patients with nAMD. The purpose of this study is to investigate the correlation of manually measured retinal fluid via conventional assessment of B-scan-related heights by certified reading center experts with fully automated AI-based quantification of retinal fluid volumes, as well as identifying the distribution of macular fluid pooling in active nAMD and disease activity using an approved and validated AI-based deep-learning algorithm. Materials and Methods This retrospective study adhered to the ethical guidelines outlined in the Declaration of Helsinki and the International Conference of Harmonization of Good Clinical Practice guidelines. The study included baseline visits of nAMD patients from multicenter study data of the Vienna Reading Center (VRC). The inclusion and exclusion criteria were similar across the studies, meaning that only treatment-naive patients with MNV secondary to AMD, without any other impairing ocular conditions, were included. Heidelberg Spectralis OCT-Volume scans with either 49 or 97 B-scans were segmented by the automatic fluid segmentation described below, and manually graded by expert readers. Human expert grading Manually measured parameters were acquired from certified readers of the VRC. The data was extracted from trial data of large randomized controlled trials for anti-VEGF agents. Highest vertical extension in the central millimeter (CMM) of subretinal fluid (SRF), intraretinal fluid (IRF), and pigment epithelium detachment (PED) was manually assessed and measured using an in-house-developed reading tool (OCTAVO). [13] Pigment epithelial detachment (PED) was defined as space between Bruch’s membrane and the outer boundary of the retinal pigment epithelium (RPE), [14] including all lesion sizes and excluding SHRM or other hyperreflective tissue on the OCT. Deep learning-based automated fluid segmentation A validated deep learning-based algorithm (RetInSight Fluid Monitor Version 2 algorithm, Vienna, Austria), CE-marked under MDR EU 2017/745 was utilized for automated fluid compartment segmentation. [12,15] The algorithm was trained on Heidelberg Spectralis OCT systems to automatically segment each OCT pixel as normal tissue, versus retinal fluid compartments including SRF, IRF, and PED. Fluid volume information was then processed, and the amount of fluid presented in nanoliters. Pixel-wise three-dimensional segmentation information is used to extract fluid volumes, including highest fluid levels in vertical extension, and their respective localization within the central 6mm subfields. Parameters collected in this way were then correlated with the parameters measured by human expert readers see above. Statistics Statistical analyses were performed using the R Statistics Version 4.1.3. For fluid level heights and volumes, the median and the interquartile range (IQR) are reported. To explore the connection between manually measured heights and those derived from the automated Fluid Monitor, Bland-Altman analysis and Pearson's correlation coefficients were calculated. R-squared values for the relationship of height vs. height measurements as well as height vs. volume measurements were calculated by linear regression. The Chi-square test was used to compare proportions of maximum fluid locations. Results Our study cohort included 890 baseline visits. In this treatment-naïve nAMD population, we identified IRF in both manually and automatically in 328 eyes, SRF in 502 eyes, and PED in 705 eyes in the CMM. Figure 1 shows OCT B-scans of nAMD patients with the respective fluid compartments segmented by an AI-based algorithm. The last row of Figure 1 shows the manual and automatic measurements of the maximum vertical fluid expansion for SRF in this case. Table 1 shows the median and interquartile range (IQR) of the manually and automatically measured macular fluid parameters. Measurements of IRF and PED by B-scan-based height were consistent between manual human readers and the automated algorithm. In contrast, for SRF, the manually acquired values were significantly higher than those obtained by the AI-based algorithm. As seen in the last row of Figure 1, the manual measurement (first column) consistently included slightly hyperreflective material in the subretinal space, while the automatic method (second column) tended to omit these hyperreflective areas. Table 1: Median (IQR) macular fluid heights and volumes Human Expert AI-based Fluid Monitoring Maxiumum height in CMM (µm) Maxiumum height in CMM (µm) p-value Volume in CMM (nl) Volume in 6MM (nl) IRF 198 (124-284) 186 (119-280) 0.45 23 (7-57) 83 (0-313) SRF 124 (74-186) 85 (46-143) <0.001 12 (2-46) 79 (4-305) PED 196 (87-236) 190 (85-236) 0.323 77 (18-94) 264 (9-267) IQR=interquartile range, CMM=central millimeter, 6MM=central 6 mm, IRF=intraretinal fluid, SRF=subretinal fluid, PED=pigment epithelium detachment Figure 2 shows Bland-Altman plots and scatter plots with linear regression lines for the comparison of manually measured maximum height in the CMM and the maximum AI-measured height in the CMM. Correlations between manually and AI-measured heights in the CMM were found to be strong for IRF r = 0.87 and PED r= 0.91, and moderate for SRF r= 0.67. Bland-Altman analysis revealed a mean error for IRF, SRF, and PED of -8.7 µm, -39.0 µm, and -6.5 µm, respectively. Upper and lower limits of agreement were 105.4 µm and -122.8µm (Figure 2A) for IRF, 97.9 µm and -175.9 µm (Figure 2C) for SRF, and 135.6µm and -148.6µm for PED (Figure 2E). R-Squared of the linear regression model for IRF (Figure 2B), SRF (Figure 2D), and PED (Figure 2F) was 0.74, 0.45, and 0.82, respectively. Highest fluid levels located by the Fluid Monitor The histogram in Figure 3 shows the distribution of fluid volumes for IRF in (A), SRF (B), PED (C) and the distribution of the corresponding locations of maximum fluid height indicated by shades of gray. Distribution of maximum fluid level locations are shown in Table 2. IRF peaked most frequently on B-scans in the CMM closely followed by the 1-3mm area. The vertical peak was significantly more often seen in the CMM and 1-3mm compared to the 3-6 mm area (both p < 0.001), but not significantly more often than between CMM and 1-3 mm. The highest SRF amount peaked significantly more frequently between 3 and 6 mm than between 1 and 3 mm (p = 0.007) and the CMM (p < 0.001). Maxima of highest vertical PED extensions were significantly more often between 1 and 3 mm than in the CMM (p < 0.001) and between 3 and 6 mm (p < 0.001). Table 2: Highest fluid level locations determined by the Fluid Monitor, n (%) IRF SRF PED n=783 n=867 n=869 CMM 300 (38) 194 (22) 211 (24) 1 - 3 mm 281 (36) 302 (35) 508 (58) 3 - 6 mm 202 (26) *CMM,1-3 371 (43) 150 (17) CMM=central millimeter, 1-3mm=ring between one to three millimeters diameter, 3-6mm=ring between three to six millimeters diameter, IRF=intraretinal fluid, SRF=subretinal fluid, PED=pigment epithelial detachment Comparison of volumes and vertical height of individual fluid compartments Figure 4 shows scatterplots with linear regression lines of manually and automatically measured maximum vertical fluid heights compared to three-dimensional fluid volumes determined by the Fluid Monitor in the CMM. Pearson correlation between manual height measurements and AI-based volumes in the CMM was strong for IRF (r=0.76), and PED (r=0.87) and moderate for SRF (r=0.55), indicating that manual fluid height measurements cannot grasp the entirety of exudative disease activity. Comparing AI-based fluid monitor maximum height with AI-based fluid volumes all fluid compartments showed a strong correlation (IRF: r=0.86, SRF: r=0.84, PED: r=0.93). Figure 5 shows the comparison between manual and AI maximum height measurements of the CMM vs. AI volumes in the central 6mm of the macula. As expected, the correlation between SRF height measurements in the CMM and the fluid volume within the central 6mm is the weakest. This indicates that, the SRF volume spreads over a flatter and broader area. However, the total PED volume in the central 6mm can still be estimated with a moderate correlation to manual vertical height measurements. For determining the IRF volume within the central 6mm, the CMM measurements were only suitable with a low to moderate correlation. Discussion The aim of our analysis was to identify the correlation and consistency between human expert-based retinal fluid assessment and automated AI-based quantification. The parameters included the vertical maximum height observed on each B-scan, which is typically used in standardized reading center settings with manual reading protocols. Additionally, we considered automated quantification of vertical B-scan extension using a validated tool and three-dimensional fluid volume determination, a classic feature of automated segmentation. The results of this head-to-head comparison will provide insights into the efficiency and accuracy of quantifying pathological retinal fluid activity in nAMD. The relationship between the human and AI maximum fluid height measurements in the central millimeter was particularly strong for IRF and PED, while the correlation for SRF was moderate. The weaker correlation for SRF can be attributed to its typically broader horizontal distribution, making it more difficult to manually identify the maximum and the mildly hyperreflective material within the subretinal space. This introduced subjectivity in determining the true boundaries of the subretinal space. According to the reading center’s protocol, human experts tended to include mildly hyperreflective material, whereas the AI-based algorithm excluded areas with even mild hyperreflectivity in the subretinal space. First, we compared the identification of B-scan-related fluid height between human capacity and automated analysis. Traditionally, large clinical trials used central retinal thickness (CRT) as a key measure to assess morphological changes after anti-VEGF treatment. [ 8 , 16 ] CRT may be considered a surrogate measure for IRF and SRF, the main indicators for anti-VEGF treatment decisions. The dilemma of with CRT is that it poorly reflects the actual total disease activity of IRF and SRF due to various influencing factors in nAMD patients, [ 10 , 11 ] such as pigment epithelial detachments (often included in CRT measurements), fibrosis, neovascular membranes, SHRM, and other morphologic pathologies. Each of these factors exhibit distinct dynamics during the treatment of the disease and in addition CRT only addresses the central millimeter of the macular. Therefore, CRT was found to correlate only poorly with objective fluid pooling in nAMD with an R < 0.5. [ 10 ] However, although the majority of MNV appear subfoveally [ 17 , 18 ], maximum vertical extensions of macular fluid compartments frequently occur outside the central millimeter. For example, it is useful to quantify fluid compartments beyond the central millimeter, in type 3 MNV lesions which are mainly located temporally parafoveally, with their maximum IRF accumulation in that area. [ 19 ] Furthermore, fluid estimation over the entire scan volumes is required in central geographic atrophies (GA) which is typically associated with thinning in the central retina and MNV may develop adjacent to the border of the GA despite the advanced stage of disease. [ 20 ] This fact will be most crucial when novel treatments might lead to an increased rate of MNV in patients with GA. [ 21 , 22 ] Manually assessing maxima of fluids over whole OCT volumes, counting 47 to 92 individual B-scans, is much too time-consuming, difficult, and always involves the subjective character of a human decision. In our analysis we highlight the consistency of the algorithmic measurement for IRF and PED, but also the inconsistency in SRF. Obviously, pixel-wise raster-based segmentation in a fast automated manner can objectively and precisely identify the relevant peaks in each B-scan at any location. In contrast, automatic AI-based fluid segmentation can provide rapid and precise quantification of fluid volumes as well as distribution of the respective fluid types in the retinal layers. As we showed in our study, only IRF most frequently had its maximum height in the central millimeter and 1-3mm (non-significant difference), while the distribution of SRF and PED showed their maxima in the parafoveal region (1-3mm). Similar results were shown for nAMD in the study of Michl et al. where the maximum for IRF was in the central millimeter, and the maximum for SRF was parafoveally, indicating the urgent need to quantify disease activity also beyond the CMM. [ 23 ] In a second step, we compared the consistency of vertical height and three-dimensional volume to provide evidence whether vertical height is a robust representation of overall fluid amount as the use of CRT suggests. The correlation between AI-based fluid monitor height measurements and AI-based fluid volumes in the central millimeter was strong for all fluid compartments, with Pearson correlation coefficients ranging from r = 0.84 to r = 0.93, but weaker for the manual measurements where correlation coefficients were r = 0.55 for SRF, r = 0.76 for IRF, and r = 0.87 for PED. Moreover, this correlation was markedly worse when comparing maximum heights in the central millimeter with total fluid volumes of the central 6mm. This suggests that manual as well as AI-based height measurements can be used to estimate AI fluid volumes in the central millimeter only but are not sufficient for estimating the actual fluid volume of the entire macular area. This implies that physicians might miss important disease activity signs by just focusing morphologically on the central millimeter. This is particularly relevant, as IRF, SRF and PED are not distributed evenly, but rather separated and counter located. [ 14 ] Using AI to assess macular fluid compartments by volume and location has the potential to significantly accelerate and objectify the assessment of fluid accumulation and dynamics in nAMD. Moreover, since this technology has the potential to accurately measure fluid activity it may serve as a VEGF-meter as proposed by Rosenfeld [ 24 ], paving the way for optimized anti-VEGF treatment regimens for each individual nAMD patient. Chakravarthy et al have already shown, that it is the fluctuation of retinal fluid volumes which reflects disease activity and causes long-term progressive vision loss. [ 25 ] Only AI-based assessment can provide such information, which is true for the vertical height as well as the total volume assessment. As AI advances, it is anticipated to also integrate novel imaging modalities, such as wide-field high resolution OCT angiography, enabling even more precise measurements and the identification of further pathological markers within the retina, which will help us to understand and treat nAMD with even greater precision. Our study has some limitations. Since are retrospective, there is no baseline data on demographics, visual function, scan quality, MNV type and location. This disadvantage did not allow us to establish correlations between certain factors of interest and the dataset might not represent all nAMD patients in clinical routine. The automated Fluid Monitor has been extensively validated and is capable of benchmarking against human fluid height measurements, which may include a subjective component, and where consensus regarding SHRM or other coexisting alterations is not always reached. The advantage of this study is the large dataset used and an unbiased approach in comparing human measurements with AI-quantified fluid using an approved fluid measurement tool. Conclusion Automated AI-based retinal fluid assessment adds another dimension to conventional manual retinal fluid height measurements, providing detailed quantitative measurements of IRF, SRF, and PED parameters to accurately grasp the full extent of disease activity in nAMD. This advantage of AI-based evaluation applies to conventional parameters such as fluid height, a major component of common CRT measurements, but even more so to identify fluid in its realistic three-dimensional extension. Automated fluid measurement offers advantages especially in areas where large amounts of data must be evaluated which in nAMD is the real-world practice, and where manual measurements are unfeasible and cannot be standardized. We therefore believe that AI fluid monitoring provides a significant contribution to the analysis and interpretation of nAMD trials and might be the base for precise personalized therapies in clinical practice. Declarations Competing Interests S.S. none.B.S.G. is a scientific consultant for Roche, Zeiss, Bayer, and Abbvie. G.D. none.O.L. none.A.W. none.H.B. received research funds from Heidelberg Engineering and Apellis.G.S.R is a scientific consultant for Apellis, Bayer, Boehringer Ingelheim and Roche and received research funds from RetInSight. U.S.-E. is a scientific consultant for Apellis, Bayer, AbbVie, Medscape, Allergan, Roche, Boehringer, Aviceda, Annexon, Topcon, Alkeus and received research funds by Genentech, Kodiak, Novartis, RetInSight, Apellis Pharmaceuticals. Ethics statement The study received approval from the Ethics Committee of the Medical University of Vienna. It was conducted in accordance with the Declaration of Helsinki. All patients provided written informed consent. Author Contribution G.S.R., U.S.-E., B.S.G., G.D., S.S. conceptualized the analysis and monitored data preparation. S.S. performed statistical analysis of the data, drafted the manuscript, and prepared the figures. O.L., A.W., and H.B. analyzed the OCT images. G.S.R., B.S.G., O.L., H.B., U.S.-E. revised the manuscript and approved final submission. Data Availability The data are available from the corresponding author upon reasonable request. References Bourne RRA, Stevens GA, White RA, Smith JL, Flaxman SR, Price H, et al. Causes of vision loss worldwide, 1990-2010: A systematic analysis. Lancet Glob Health. 2013 Dec;1(6). Pennington KL, DeAngelis MM. Epidemiology of age-related macular degeneration (AMD): associations with cardiovascular disease phenotypes and lipid factors. Vol. 3, Eye and Vision. BioMed Central Ltd; 2016. Wong WL, Su X, Li X, Cheung CMG, Klein R, Cheng CY, et al. Global prevalence of age-related macular degeneration and disease burden projection for 2020 and 2040: A systematic review and meta-analysis. Lancet Glob Health. 2014 Feb;2(2). Spaide RF, Jaffe GJ, Sarraf D, Freund KB, Sadda SR, Staurenghi G, et al. Consensus Nomenclature for Reporting Neovascular Age-Related Macular Degeneration Data: Consensus on Neovascular Age-Related Macular Degeneration Nomenclature Study Group. Ophthalmology. 2020 May 1;127(5):616–36. Schmidt-Erfurth U, Waldstein SM. A paradigm shift in imaging biomarkers in neovascular age-related macular degeneration. Vol. 50, Progress in Retinal and Eye Research. Elsevier Ltd; 2016. p. 1–24. Schmidt-Erfurth U, Vogl WD, Jampol LM, Bogunović H. Application of Automated Quantification of Fluid Volumes to Anti–VEGF Therapy of Neovascular Age-Related Macular Degeneration. Ophthalmology. 2020 Sep 1;127(9):1211–9. Berta A, Larsen M, Macfadden W. Treat-and-Extend versus Monthly Regimen in Neovascular Age-Related Macular Degeneration Results with Ranibizumab from the TREND Study. 2017;57–65. Busbee BG, Ho AC, Brown DM, Heier JS, Suñer IJ. Twelve-Month Efficacy and Safety of 0 . 5 mg or 2 . 0 mg Ranibizumab in Patients with Subfoveal Neovascular Age-related Macular. 2013;1046–56. Dugel PU, Koh A, Ogura Y, Jaffe GJ, Schmidt-Erfurth U, Brown DM, et al. HAWK and HARRIER: Phase 3, Multicenter, Randomized, Double-Masked Trials of Brolucizumab for Neovascular Age-Related Macular Degeneration. In: Ophthalmology. Elsevier Inc.; 2020. p. 72–84. Pawloff M, Bogunovic H, Gruber A, Michl M, Riedl S, Schmidt-Erfurth U. SYSTEMATIC CORRELATION OF CENTRAL SUBFIELD THICKNESS WITH RETINAL FLUID VOLUMES QUANTIFIED BY DEEP LEARNING IN THE MAJOR EXUDATIVE MACULAR DISEASES [Internet]. Available from: http://links.lww.com/IAE/ Pawloff M, Gerendas BS, Deak G, Bogunovic H, Gruber A, Schmidt-Erfurth U. Performance of retinal fluid monitoring in OCT imaging by automated deep learning versus human expert grading in neovascular AMD. Eye (Basingstoke). 2023;(December 2022):1–8. Schlegl T, Waldstein SM, Bogunovic H, Endstraßer F, Sadeghipour A, Philip AM, et al. Fully Automated Detection and Quantification of Macular Fluid in OCT Using Deep Learning. Ophthalmology [Internet]. 2018;125(4):549–58. Available from: https://doi.org/10.1016/j.ophtha.2017.10.031 Waldstein SM, Gerendas BS, Montuoro A, Simader C, Schmidt-Erfurth U. Quantitative comparison of macular segmentation performance using identical retinal regions across multiple spectral-domain optical coherence tomography instruments. British Journal of Ophthalmology. 2015 Jun 1;99(6):794–800. Klimscha S, Waldstein SM, Schlegl T, Bogunović H, Sadeghipour A, Philip AM, et al. Spatial correspondence between intraretinal fluid, subretinal fluid, and pigment epithelial detachment in neovascular age-related macular degeneration. Invest Ophthalmol Vis Sci. 2017 Aug 1;58(10):4039–48. Schmidt-Erfurth U, Reiter GS, Riedl S, Seeböck P, Vogl WD, Blodi BA, et al. AI-based monitoring of retinal fluid in disease activity and under therapy. Vol. 86, Progress in Retinal and Eye Research. Elsevier Ltd; 2022. Berta A, Larsen M, Macfadden W. Treat-and-Extend versus Monthly Regimen in Neovascular Age-Related Macular Degeneration Results with Ranibizumab from the TREND Study. 2017;57–65. Cohen SY, Creuzot-Garcher C, Darmon J, Desmettre T, Korobelnik JF, Levrat F, et al. Types of choroidal neovascularisation in newly diagnosed exudative age-related macular degeneration. British Journal of Ophthalmology. 2007;91(9):1173–6. Olsen TW, Feng X, Kasper TJ, Rath PP, Steuer ER, Kertes PJ. Fluorescein angiographic lesion type frequency in neovascular age-related macular degeneration. Vol. 5, Evidence-Based Eye Care. 2004. p. 167–9. Haj Najeeb B, Deak G, Schmidt-erfurth U, Gerendas BS. THE RAP STUDY, REPORT TWO The Regional Distribution of Macular Neovascularization Type 3, a Novel Insight Into Its Etiology [Internet]. Available from: http://journals.lww.com/retinajournal Mettu PS, Allingham MJ, Cousins SW. Incomplete response to Anti-VEGF therapy in neovascular AMD: Exploring disease mechanisms and therapeutic opportunities. Vol. 82, Progress in Retinal and Eye Research. Elsevier Ltd; 2021. Heier JS, Lad EM, Holz FG, Rosenfeld PJ, Guymer RH, Boyer D, et al. Articles Pegcetacoplan for the treatment of geographic atrophy secondary to age-related macular degeneration (OAKS and DERBY): two multicentre, randomised, double-masked, sham-controlled, phase 3 trials [Internet]. Vol. 402, www.thelancet.com. 2023. Available from: www.thelancet.com Khanani AM, Patel SS, Staurenghi G, Tadayoni R, Danzig CJ, Eichenbaum DA, et al. Efficacy and safety of avacincaptad pegol in patients with geographic atrophy (GATHER2): 12-month results from a randomised, double-masked, phase 3 trial. The Lancet. 2023 Oct; Michl M, Fabianska M, Seeböck P, Sadeghipour A, Haj Najeeb B, Bogunovic H, et al. Automated quantification of macular fluid in retinal diseases and their response to anti-VEGF therapy. British Journal of Ophthalmology. 2022;106(1):113–20. Rosenfeld PJ. Optical coherence tomography and the development of antiangiogenic therapies in neovascular age-related macular degeneration. Invest Ophthalmol Vis Sci. 2016;57(9):OCT14–26. Chakravarthy U, Havilio M, Syntosi A, Pillai N, Wilkes E, Benyamini G, et al. Impact of macular fluid volume fluctuations on visual acuity during anti-VEGF therapy in eyes with nAMD. Eye (Basingstoke). 2021 Nov 1;35(11):2983–90. Additional Declarations Competing interest reported. S.S. none. B.S.G. is a scientific consultant for Roche, Zeiss, Bayer, and Abbvie. G.D. none. O.L. none. A.W. none. H.B. received research funds from Heidelberg Engineering and Apellis. G.S.R is a scientific consultant for Apellis, Bayer, Boehringer Ingelheim and Roche and received research funds from RetInSight. U.S.-E. is a scientific consultant for Apellis, Bayer, AbbVie, Medscape, Allergan, Roche, Boehringer, Aviceda, Annexon, Topcon, Alkeus and received research funds by Genentech, Kodiak, Novartis, RetInSight, Apellis Pharmaceuticals. Cite Share Download PDF Status: Published Journal Publication published 31 Mar, 2026 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 19 Dec, 2025 Reviews received at journal 19 Jun, 2025 Reviewers agreed at journal 14 Jun, 2025 Reviewers agreed at journal 12 Jun, 2025 Reviews received at journal 11 Jan, 2025 Reviewers agreed at journal 27 Dec, 2024 Reviewers invited by journal 10 Nov, 2024 Editor assigned by journal 02 Nov, 2024 Editor invited by journal 25 Oct, 2024 Submission checks completed at journal 23 Oct, 2024 First submitted to journal 22 Oct, 2024 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5313889","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":373762136,"identity":"a8ee18ba-c259-47ed-91f9-06631af42a93","order_by":0,"name":"Stefan Steiner","email":"","orcid":"","institution":"Medical University of Vienna","correspondingAuthor":false,"prefix":"","firstName":"Stefan","middleName":"","lastName":"Steiner","suffix":""},{"id":373762137,"identity":"d7f3a32f-b51a-4cf8-99af-cb859f513f15","order_by":1,"name":"Bianca S. Gerendas","email":"","orcid":"","institution":"Medical University of Vienna","correspondingAuthor":false,"prefix":"","firstName":"Bianca","middleName":"S.","lastName":"Gerendas","suffix":""},{"id":373762138,"identity":"42ab7e5c-1e35-4d43-a544-12d97436dcf4","order_by":2,"name":"Gabor Deak","email":"","orcid":"","institution":"Medical University of Vienna","correspondingAuthor":false,"prefix":"","firstName":"Gabor","middleName":"","lastName":"Deak","suffix":""},{"id":373762139,"identity":"38212c91-bdd6-4915-b510-f02c99f32bf2","order_by":3,"name":"Oliver Leingang","email":"","orcid":"","institution":"RetInSight","correspondingAuthor":false,"prefix":"","firstName":"Oliver","middleName":"","lastName":"Leingang","suffix":""},{"id":373762140,"identity":"93f31f75-f92c-4199-9599-ae41f59db6b8","order_by":4,"name":"Ariadne Whitby","email":"","orcid":"","institution":"RetInSight","correspondingAuthor":false,"prefix":"","firstName":"Ariadne","middleName":"","lastName":"Whitby","suffix":""},{"id":373762141,"identity":"75770bca-6aec-404d-a3c2-f8321a46fa9d","order_by":5,"name":"Hrvoje Bogunovic","email":"","orcid":"","institution":"Medical University of Vienna","correspondingAuthor":false,"prefix":"","firstName":"Hrvoje","middleName":"","lastName":"Bogunovic","suffix":""},{"id":373762142,"identity":"7cac644e-dacc-45e7-8b08-b2d3effaf815","order_by":6,"name":"Gregor S. Reiter","email":"","orcid":"","institution":"Medical University of Vienna","correspondingAuthor":false,"prefix":"","firstName":"Gregor","middleName":"S.","lastName":"Reiter","suffix":""},{"id":373762143,"identity":"0d457cf0-7052-4bad-9611-fc31157e017d","order_by":7,"name":"Ursula Schmidt-Erfurth","email":"data:image/png;base64,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","orcid":"","institution":"Medical University of Vienna","correspondingAuthor":true,"prefix":"","firstName":"Ursula","middleName":"","lastName":"Schmidt-Erfurth","suffix":""}],"badges":[],"createdAt":"2024-10-22 18:23:12","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5313889/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5313889/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-026-44982-8","type":"published","date":"2026-03-31T15:59:43+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":68691371,"identity":"80817d6e-5abe-4bc2-bfba-9360aa578409","added_by":"auto","created_at":"2024-11-11 05:59:26","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1090390,"visible":true,"origin":"","legend":"\u003cp\u003eOCT B-scans of with and without AI-based retinal fluid segmentation. The first column shows exemplary OCT B-scans of the macula in neovascular AMD patients. The second column shows the scans with the AI-based segmentation of the intraretinal fluid (IRF) in red, the subretinal fluid (SRF) in yellow and pigment epithelial detachment (PED) in pale blue. The first row shows the different fluid volumes and distribution of a macular OCT B-scan. The second row shows IRF with large vertical extension. Row three shows a B-scan with high PED volume. Row four shows the difference in the maximum vertical expansion of SRF when compared manually (first column - higher value) to AI-based assessment (second column - lower value). The reason for the differences is mildly hyperreflective material in the subretinal space, which was often included in manual measurements but frequently omitted by AI.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-5313889/v1/15f87dacbb5456a836471711.png"},{"id":68691367,"identity":"2a25d44b-c60c-47bd-a239-409e04f026dc","added_by":"auto","created_at":"2024-11-11 05:59:26","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":201314,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of manually vs. automatically measured maximum fluid heights in the central millimeter (CMM); The first column shows Bland-Altman plots and second column scatterplots for IRF (A-B), SRF (C-D), PED (E-F).\u003c/p\u003e","description":"","filename":"Fig2.png","url":"https://assets-eu.researchsquare.com/files/rs-5313889/v1/e7bed9ee47840e52efb7f56b.png"},{"id":68691370,"identity":"8b598dd9-6d9e-4778-a647-381841239e43","added_by":"auto","created_at":"2024-11-11 05:59:26","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":46125,"visible":true,"origin":"","legend":"\u003cp\u003eHistograms showing the distribution of fluid volumes for IRF (A), SRF (B), PED (C) and the distribution of the corresponding locations of maximum fluid compartment height indicated by shades of gray. Bin width for IRF= 50nl, SRF= 85nl, PED= 280nl.\u003c/p\u003e","description":"","filename":"Fig3.png","url":"https://assets-eu.researchsquare.com/files/rs-5313889/v1/535711c7ac7b567527a2a869.png"},{"id":68691694,"identity":"22fdd781-6de8-45c7-8b34-22220c844684","added_by":"auto","created_at":"2024-11-11 06:07:26","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":409997,"visible":true,"origin":"","legend":"\u003cp\u003eScatterplots with linear regression line of manually measured (A, C, E) and AI maximum fluid height (B, D, F) vs. AI fluid volume in the CMM for IRF (A-B), SRF, (C-D), and PED (E-F).\u003c/p\u003e","description":"","filename":"Fig4.png","url":"https://assets-eu.researchsquare.com/files/rs-5313889/v1/4847f5db961db3786ff89ebd.png"},{"id":68691368,"identity":"0b0c30fc-cd38-4975-9de3-d9e97e2b34ef","added_by":"auto","created_at":"2024-11-11 05:59:26","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":423593,"visible":true,"origin":"","legend":"\u003cp\u003eScatterplots with linear regression line of manually measured (A, C, E) and AI maximum fluid height (B, D, F) in the CMM vs. AI fluid volumes of the central 6mm for IRF (A-B), SRF, (C-D), and PED (E-F).\u003c/p\u003e","description":"","filename":"Fig5.png","url":"https://assets-eu.researchsquare.com/files/rs-5313889/v1/ff8dd54c3f682e5d813ec8fb.png"},{"id":106343593,"identity":"84ee7abe-716c-42a1-b660-dedd5b782c9f","added_by":"auto","created_at":"2026-04-07 16:06:53","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2914317,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5313889/v1/9ee3d595-4ddb-44ee-8c2d-8da8b74b8ba5.pdf"}],"financialInterests":"Competing interest reported. S.S. none.\nB.S.G. is a scientific consultant for Roche, Zeiss, Bayer, and Abbvie. \nG.D. none.\nO.L. none.\nA.W. none.\nH.B. received research funds from Heidelberg Engineering and Apellis.\nG.S.R is a scientific consultant for Apellis, Bayer, Boehringer Ingelheim and Roche and received research funds from RetInSight. \nU.S.-E. is a scientific consultant for Apellis, Bayer, AbbVie, Medscape, Allergan, Roche, Boehringer, Aviceda, Annexon, Topcon, Alkeus and received research funds by Genentech, Kodiak, Novartis, RetInSight, Apellis Pharmaceuticals.","formattedTitle":"Correlation between human expert macular fluid height assessment and fluid volume quantification in neovascular age-related macular degeneration","fulltext":[{"header":"Introduction","content":"\u003cp\u003eAge-related macular degeneration (AMD) is a widespread and debilitating ocular condition primarily affecting the elderly, with a growing impact not only on patients but also on ophthalmologists, insurance companies, and healthcare providers, due to increased life expectancy.[\u003cspan additionalcitationids=\"CR2\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e] Neovascular AMD (nAMD) is characterized by an increased release of pro-angiogenic factors such as vascular endothelial growth factor (VEGF), leading to macular neovascularization (MNV) resulting in the accumulation of fluid in different retinal layers which causes vision impairment and distortion. Gold-standard for treating nAMD is the application of intravitreal anti-VEGF, with the aim to reduce the amount and recurrence of retinal fluid and prevent irreversible scaring of the macula. Accumulation of intraretinal (IRF) and subretinal fluid (SRF) are the main drivers in clinical decision-making whether a patient needs anti-VEGF administration.[\u003cspan additionalcitationids=\"CR5\" citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e] With optical coherence tomography (OCT) as advanced retinal imaging technology macular fluid compartments like IRF, SRF, and pigment epithelium detachment (PED) can be visualized in high resolution. Nevertheless, so far it has been time-consuming and challenging to quantify the amount of three-dimensional retinal fluids present in volumes rather than individual B-scans. Manual reading center measurements (e.g., central retinal thickness, central subfield thickness, and center point thickness) have historically been used as surrogate markers for the accumulation of retinal fluids, disease activity and were the base for dosing regimen in clinical trials.[\u003cspan additionalcitationids=\"CR8\" citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e] These semi-quantitative indicators were introduced since time-domain OCT lacked the raster scanning capability of today\u0026rsquo;s OCT technology not able to perform fluid quantifications throughout the entire volume. In addition, available evidence suggests that these semi-quantitative measurements are poor indicators of total disease activity.[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]\u003c/p\u003e \u003cp\u003eArtificial intelligence (AI) algorithms can precisely identify and segment individual fluid compartments in nAMD from entire OCT volumes in a fraction of time compared to a human counting fluid associated pixels, on high numbers of B-scans, allowing for the detection and quantification of nAMD-related exudation. [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e] Visualization and quantification of AI-based features support ophthalmologists in detecting and monitoring retinal fluid over time, enabling personalized treatment and efficient therapeutic interventions. By automating retinal segmentation, AI reduces the burden on healthcare professionals, enhances diagnostic precision, while improving workflows and potentially enhancing patient care and visual outcomes for patients with nAMD.\u003c/p\u003e \u003cp\u003eThe purpose of this study is to investigate the correlation of manually measured retinal fluid via conventional assessment of B-scan-related heights by certified reading center experts with fully automated AI-based quantification of retinal fluid volumes, as well as identifying the distribution of macular fluid pooling in active nAMD and disease activity using an approved and validated AI-based deep-learning algorithm.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cp\u003eThis retrospective study adhered to the ethical guidelines outlined in the Declaration of Helsinki and the International Conference of Harmonization of Good Clinical Practice guidelines. The study included baseline visits of nAMD patients from multicenter study data of the Vienna Reading Center (VRC). The inclusion and exclusion criteria were similar across the studies, meaning that only treatment-naive patients with MNV secondary to AMD, without any other impairing ocular conditions, were included. Heidelberg Spectralis OCT-Volume scans with either 49 or 97 B-scans were segmented by the automatic fluid segmentation described below, and manually graded by expert readers.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eHuman expert grading\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eManually measured parameters were acquired from certified readers of the VRC. The data was extracted from trial data of large randomized controlled trials for anti-VEGF agents. Highest vertical extension in the central millimeter (CMM) of subretinal fluid (SRF), intraretinal fluid (IRF), and pigment epithelium detachment (PED) was manually assessed and measured using an in-house-developed reading tool (OCTAVO). [13]\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003ePigment epithelial detachment (PED) was defined as space between Bruch\u0026rsquo;s membrane and the outer boundary of the retinal pigment epithelium (RPE),\u0026nbsp;[14] including all lesion sizes and excluding SHRM or other hyperreflective tissue on the OCT.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDeep learning-based automated fluid segmentation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA validated deep learning-based algorithm (RetInSight Fluid Monitor Version 2 algorithm, Vienna, Austria), CE-marked under MDR EU 2017/745 was utilized for automated fluid compartment segmentation.\u0026nbsp;[12,15]\u0026nbsp;The algorithm was trained on Heidelberg Spectralis OCT systems to automatically segment each OCT pixel as normal tissue, versus retinal fluid compartments including SRF, IRF, and PED. Fluid volume information was then processed, and the amount of fluid presented in nanoliters. Pixel-wise three-dimensional segmentation information is used to extract fluid volumes, including highest fluid levels in vertical extension, and their respective localization within the central 6mm subfields. Parameters collected in this way were then correlated with the parameters measured by human expert readers see above.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStatistics\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eStatistical analyses were performed using the R Statistics Version 4.1.3. For fluid level heights and volumes, the median and the interquartile range (IQR) are reported. To explore the connection between manually measured heights and those derived from the automated Fluid Monitor, Bland-Altman analysis and Pearson\u0026apos;s correlation coefficients were calculated. R-squared values for the relationship of height vs. height measurements as well as height vs. volume measurements were calculated by linear regression. The\u003c/p\u003e\n\u003cp\u003eChi-square test was used to compare proportions of maximum fluid locations.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eOur study cohort included 890 baseline visits. In this treatment-na\u0026iuml;ve nAMD population, we identified IRF in both manually and automatically in 328 eyes, SRF in 502 eyes, and PED in 705 eyes in the CMM.\u0026nbsp;\u003cbr\u003e\u0026nbsp;Figure 1 shows OCT B-scans of nAMD patients with the respective fluid compartments segmented by an AI-based algorithm. The last row of Figure 1 shows the manual and automatic measurements of the maximum vertical fluid expansion for SRF in this case. Table 1 shows the median and interquartile range (IQR) of the manually and automatically measured macular fluid parameters. Measurements of IRF and PED by B-scan-based height were consistent between manual human readers and the automated algorithm. In contrast, for SRF, the manually acquired values were significantly higher than those obtained by the AI-based algorithm. As seen in the last row of Figure 1, the manual measurement (first column) consistently included slightly hyperreflective material in the subretinal space, while the automatic method (second column) tended to omit these hyperreflective areas.\u0026nbsp;\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"652\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"6\" valign=\"bottom\" style=\"width: 99.8466%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTable 1:\u003c/strong\u003e Median (IQR) macular fluid heights and volumes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 3.2183%;\"\u003e\u003c/td\u003e\n \u003ctd style=\"width: 7.5658%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHuman Expert\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"4\" style=\"width: 25.464%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAI-based Fluid Monitoring\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 3.2183%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 7.5658%;\"\u003e\n \u003cp\u003eMaxiumum height\u003cbr\u003e\u0026nbsp; in CMM (\u0026micro;m)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.1304%;\"\u003e\n \u003cp\u003eMaxiumum height\u003cbr\u003e\u0026nbsp; in CMM (\u0026micro;m)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 4.7992%;\"\u003e\n \u003cp\u003ep-value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6.0978%;\"\u003e\n \u003cp\u003eVolume\u0026nbsp;\u003cbr\u003e\u0026nbsp; in CMM (nl)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6.4366%;\"\u003e\n \u003cp\u003eVolume\u0026nbsp;\u003cbr\u003e\u0026nbsp;in 6MM (nl)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 3.2183%;\"\u003e\n \u003cp\u003eIRF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 7.5658%;\"\u003e\n \u003cp\u003e198 (124-284)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.1304%;\"\u003e\n \u003cp\u003e186 (119-280)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 4.7992%;\"\u003e\n \u003cp\u003e0.45\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6.0978%;\"\u003e\n \u003cp\u003e23 (7-57)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6.4366%;\"\u003e\n \u003cp\u003e83 (0-313)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 3.2183%;\"\u003e\n \u003cp\u003eSRF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 7.5658%;\"\u003e\n \u003cp\u003e124 (74-186)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.1304%;\"\u003e\n \u003cp\u003e85 (46-143)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 4.7992%;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6.0978%;\"\u003e\n \u003cp\u003e12 (2-46)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6.4366%;\"\u003e\n \u003cp\u003e79 (4-305)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 3.2183%;\"\u003e\n \u003cp\u003ePED\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 7.5658%;\"\u003e\n \u003cp\u003e196 (87-236)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8.1304%;\"\u003e\n \u003cp\u003e190 (85-236)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 4.7992%;\"\u003e\n \u003cp\u003e0.323\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6.0978%;\"\u003e\n \u003cp\u003e77 (18-94)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6.4366%;\"\u003e\n \u003cp\u003e264 (9-267)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"6\" style=\"width: 36.2481%;\"\u003e\n \u003cp\u003eIQR=interquartile range, CMM=central millimeter, 6MM=central 6 mm, IRF=intraretinal fluid, SRF=subretinal fluid, PED=pigment epithelium detachment\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eFigure 2 shows Bland-Altman plots and scatter plots with linear regression lines for the comparison of manually measured maximum height in the CMM and the maximum AI-measured height in the CMM. Correlations between manually and AI-measured heights in the CMM were found to be strong for IRF r = 0.87 and PED r= 0.91, and moderate for SRF r= 0.67. Bland-Altman analysis revealed a mean error for IRF, SRF, and PED of -8.7 \u0026micro;m, -39.0 \u0026micro;m, and -6.5 \u0026micro;m, respectively. Upper and lower limits of agreement were 105.4 \u0026micro;m and -122.8\u0026micro;m (Figure 2A) for IRF, 97.9 \u0026micro;m and -175.9 \u0026micro;m (Figure 2C) for SRF, and 135.6\u0026micro;m and -148.6\u0026micro;m for PED (Figure 2E). R-Squared of the linear regression model for IRF (Figure 2B), SRF (Figure 2D), and PED (Figure 2F) was 0.74, 0.45, and 0.82, respectively.\u003c/p\u003e\n\u003ch2\u003eHighest fluid levels located by the Fluid Monitor\u003c/h2\u003e\n\u003cp\u003eThe histogram in Figure 3 shows the distribution of fluid volumes for IRF in (A), SRF (B), PED (C) and the distribution of the corresponding locations of maximum fluid height indicated by shades of gray. Distribution of maximum fluid level locations are shown in Table 2.\u0026nbsp;IRF peaked most frequently on B-scans in the CMM closely followed by the 1-3mm area. The vertical peak was significantly more often seen in the CMM and 1-3mm compared to the 3-6 mm area (both p \u0026lt; 0.001), but not significantly more often than between CMM and 1-3 mm. The highest SRF amount peaked significantly more frequently between 3 and 6 mm than between 1 and 3 mm (p = 0.007) and the CMM (p \u0026lt; 0.001). Maxima of highest vertical PED extensions were significantly more often between 1 and 3 mm than in the CMM (p \u0026lt; 0.001) and between 3 and 6 mm (p \u0026lt; 0.001).\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"376\" style=\"margin-right: calc(71%); width: 29%;\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\" valign=\"bottom\" style=\"width: 40%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTable 2:\u003c/strong\u003e Highest fluid level locations determined\u003cbr\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; by the Fluid Monitor, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 9.1892%;\"\u003e\u003c/td\u003e\n \u003ctd style=\"width: 14.8108%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eIRF\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSRF\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8%;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePED\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 9.1892%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.8108%;\"\u003e\n \u003cp\u003en=783\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8%;\"\u003e\n \u003cp\u003en=867\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8%;\"\u003e\n \u003cp\u003en=869\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 9.1892%;\"\u003e\n \u003cp\u003eCMM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.8108%;\"\u003e\n \u003cp\u003e300 (38)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8%;\"\u003e\n \u003cp\u003e194 (22)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8%;\"\u003e\n \u003cp\u003e211 (24)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 9.1892%;\"\u003e\n \u003cp\u003e1 - 3 mm\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.8108%;\"\u003e\n \u003cp\u003e281 (36)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8%;\"\u003e\n \u003cp\u003e302 (35)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8%;\"\u003e\n \u003cp\u003e508 (58)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 9.1892%;\"\u003e\n \u003cp\u003e3 - 6 mm\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.8108%;\"\u003e\n \u003cp\u003e202 (26)\u003csup\u003e*CMM,1-3\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8%;\"\u003e\n \u003cp\u003e371 (43)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8%;\"\u003e\n \u003cp\u003e150 (17)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\" style=\"width: 40%;\"\u003e\n \u003cp\u003eCMM=central millimeter, 1-3mm=ring between one to three millimeters diameter, 3-6mm=ring between three to six millimeters diameter, IRF=intraretinal fluid, SRF=subretinal fluid, PED=pigment epithelial detachment\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003ch2\u003eComparison of volumes and vertical height of individual fluid compartments\u003c/h2\u003e\n\u003cp\u003eFigure 4 shows scatterplots with linear regression lines of manually and automatically measured maximum vertical fluid heights compared to three-dimensional fluid volumes determined by the Fluid Monitor in the CMM. Pearson correlation between manual height measurements and AI-based volumes in the CMM was strong for IRF (r=0.76), and PED (r=0.87) and moderate for SRF (r=0.55), indicating that manual fluid height measurements cannot grasp the entirety of exudative disease activity. Comparing AI-based fluid monitor maximum height with AI-based fluid volumes all fluid compartments showed a strong correlation (IRF: r=0.86, SRF: r=0.84, PED: r=0.93).\u003c/p\u003e\n\u003cp\u003eFigure 5 shows the comparison between manual and AI maximum height measurements of the CMM vs. AI volumes in the central 6mm of the macula. As expected, the correlation between SRF height measurements in the CMM and the fluid volume within the central 6mm is the weakest. This indicates that, the SRF volume spreads over a flatter and broader area. However, the total PED volume in the central 6mm can still be estimated with a moderate correlation to manual vertical height measurements. For determining the IRF volume within the central 6mm, the CMM measurements were only suitable with a low to moderate correlation.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe aim of our analysis was to identify the correlation and consistency between human expert-based retinal fluid assessment and automated AI-based quantification. The parameters included the vertical maximum height observed on each B-scan, which is typically used in standardized reading center settings with manual reading protocols. Additionally, we considered automated quantification of vertical B-scan extension using a validated tool and three-dimensional fluid volume determination, a classic feature of automated segmentation. The results of this head-to-head comparison will provide insights into the efficiency and accuracy of quantifying pathological retinal fluid activity in nAMD.\u003c/p\u003e \u003cp\u003eThe relationship between the human and AI maximum fluid height measurements in the central millimeter was particularly strong for IRF and PED, while the correlation for SRF was moderate. The weaker correlation for SRF can be attributed to its typically broader horizontal distribution, making it more difficult to manually identify the maximum and the mildly hyperreflective material within the subretinal space. This introduced subjectivity in determining the true boundaries of the subretinal space. According to the reading center\u0026rsquo;s protocol, human experts tended to include mildly hyperreflective material, whereas the AI-based algorithm excluded areas with even mild hyperreflectivity in the subretinal space.\u003c/p\u003e \u003cp\u003eFirst, we compared the identification of B-scan-related fluid height between human capacity and automated analysis. Traditionally, large clinical trials used central retinal thickness (CRT) as a key measure to assess morphological changes after anti-VEGF treatment. [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e] CRT may be considered a surrogate measure for IRF and SRF, the main indicators for anti-VEGF treatment decisions. The dilemma of with CRT is that it poorly reflects the actual total disease activity of IRF and SRF due to various influencing factors in nAMD patients, [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e] such as pigment epithelial detachments (often included in CRT measurements), fibrosis, neovascular membranes, SHRM, and other morphologic pathologies. Each of these factors exhibit distinct dynamics during the treatment of the disease and in addition CRT only addresses the central millimeter of the macular. Therefore, CRT was found to correlate only poorly with objective fluid pooling in nAMD with an R\u0026thinsp;\u0026lt;\u0026thinsp;0.5. [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e] However, although the majority of MNV appear subfoveally [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e], maximum vertical extensions of macular fluid compartments frequently occur outside the central millimeter. For example, it is useful to quantify fluid compartments beyond the central millimeter, in type 3 MNV lesions which are mainly located temporally parafoveally, with their maximum IRF accumulation in that area. [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e] Furthermore, fluid estimation over the entire scan volumes is required in central geographic atrophies (GA) which is typically associated with thinning in the central retina and MNV may develop adjacent to the border of the GA despite the advanced stage of disease. [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e] This fact will be most crucial when novel treatments might lead to an increased rate of MNV in patients with GA. [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/p\u003e \u003cp\u003eManually assessing maxima of fluids over whole OCT volumes, counting 47 to 92 individual B-scans, is much too time-consuming, difficult, and always involves the subjective character of a human decision.\u003c/p\u003e \u003cp\u003eIn our analysis we highlight the consistency of the algorithmic measurement for IRF and PED, but also the inconsistency in SRF. Obviously, pixel-wise raster-based segmentation in a fast automated manner can objectively and precisely identify the relevant peaks in each B-scan at any location.\u003c/p\u003e \u003cp\u003eIn contrast, automatic AI-based fluid segmentation can provide rapid and precise quantification of fluid volumes as well as distribution of the respective fluid types in the retinal layers. As we showed in our study, only IRF most frequently had its maximum height in the central millimeter and 1-3mm (non-significant difference), while the distribution of SRF and PED showed their maxima in the parafoveal region (1-3mm). Similar results were shown for nAMD in the study of Michl et al. where the maximum for IRF was in the central millimeter, and the maximum for SRF was parafoveally, indicating the urgent need to quantify disease activity also beyond the CMM. [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]\u003c/p\u003e \u003cp\u003eIn a second step, we compared the consistency of vertical height and three-dimensional volume to provide evidence whether vertical height is a robust representation of overall fluid amount as the use of CRT suggests. The correlation between AI-based fluid monitor height measurements and AI-based fluid volumes in the central millimeter was strong for all fluid compartments, with Pearson correlation coefficients ranging from r\u0026thinsp;=\u0026thinsp;0.84 to r\u0026thinsp;=\u0026thinsp;0.93, but weaker for the manual measurements where correlation coefficients were r\u0026thinsp;=\u0026thinsp;0.55 for SRF, r\u0026thinsp;=\u0026thinsp;0.76 for IRF, and r\u0026thinsp;=\u0026thinsp;0.87 for PED. Moreover, this correlation was markedly worse when comparing maximum heights in the central millimeter with total fluid volumes of the central 6mm. This suggests that manual as well as AI-based height measurements can be used to estimate AI fluid volumes in the central millimeter only but are not sufficient for estimating the actual fluid volume of the entire macular area. This implies that physicians might miss important disease activity signs by just focusing morphologically on the central millimeter. This is particularly relevant, as IRF, SRF and PED are not distributed evenly, but rather separated and counter located. [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/p\u003e \u003cp\u003eUsing AI to assess macular fluid compartments by volume and location has the potential to significantly accelerate and objectify the assessment of fluid accumulation and dynamics in nAMD. Moreover, since this technology has the potential to accurately measure fluid activity it may serve as a VEGF-meter as proposed by Rosenfeld [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e], paving the way for optimized anti-VEGF treatment regimens for each individual nAMD patient. Chakravarthy et al have already shown, that it is the fluctuation of retinal fluid volumes which reflects disease activity and causes long-term progressive vision loss. [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e] Only AI-based assessment can provide such information, which is true for the vertical height as well as the total volume assessment. As AI advances, it is anticipated to also integrate novel imaging modalities, such as wide-field high resolution OCT angiography, enabling even more precise measurements and the identification of further pathological markers within the retina, which will help us to understand and treat nAMD with even greater precision.\u003c/p\u003e \u003cp\u003eOur study has some limitations. Since are retrospective, there is no baseline data on demographics, visual function, scan quality, MNV type and location. This disadvantage did not allow us to establish correlations between certain factors of interest and the dataset might not represent all nAMD patients in clinical routine. The automated Fluid Monitor has been extensively validated and is capable of benchmarking against human fluid height measurements, which may include a subjective component, and where consensus regarding SHRM or other coexisting alterations is not always reached. The advantage of this study is the large dataset used and an unbiased approach in comparing human measurements with AI-quantified fluid using an approved fluid measurement tool.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eAutomated AI-based retinal fluid assessment adds another dimension to conventional manual retinal fluid height measurements, providing detailed quantitative measurements of IRF, SRF, and PED parameters to accurately grasp the full extent of disease activity in nAMD. This advantage of AI-based evaluation applies to conventional parameters such as fluid height, a major component of common CRT measurements, but even more so to identify fluid in its realistic three-dimensional extension. Automated fluid measurement offers advantages especially in areas where large amounts of data must be evaluated which in nAMD is the real-world practice, and where manual measurements are unfeasible and cannot be standardized. We therefore believe that AI fluid monitoring provides a significant contribution to the analysis and interpretation of nAMD trials and might be the base for precise personalized therapies in clinical practice.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eCompeting Interests\u003c/h2\u003e\u003cp\u003eS.S. none.B.S.G. is a scientific consultant for Roche, Zeiss, Bayer, and Abbvie. G.D. none.O.L. none.A.W. none.H.B. received research funds from Heidelberg Engineering and Apellis.G.S.R is a scientific consultant for Apellis, Bayer, Boehringer Ingelheim and Roche and received research funds from RetInSight. U.S.-E. is a scientific consultant for Apellis, Bayer, AbbVie, Medscape, Allergan, Roche, Boehringer, Aviceda, Annexon, Topcon, Alkeus and received research funds by Genentech, Kodiak, Novartis, RetInSight, Apellis Pharmaceuticals.\u003c/p\u003e\u003c/p\u003e\u003cp\u003e \u003ch2\u003eEthics statement\u003c/h2\u003e \u003cp\u003e The study received approval from the Ethics Committee of the Medical University of Vienna. It was conducted in accordance with the Declaration of Helsinki. All patients provided written informed consent.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eG.S.R., U.S.-E., B.S.G., G.D., S.S. conceptualized the analysis and monitored data preparation. S.S. performed statistical analysis of the data, drafted the manuscript, and prepared the figures. O.L., A.W., and H.B. analyzed the OCT images. G.S.R., B.S.G., O.L., H.B., U.S.-E. revised the manuscript and approved final submission.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe data are available from the corresponding author upon reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eBourne RRA, Stevens GA, White RA, Smith JL, Flaxman SR, Price H, et al. Causes of vision loss worldwide, 1990-2010: A systematic analysis. Lancet Glob Health. 2013 Dec;1(6). \u003c/li\u003e\n\u003cli\u003ePennington KL, DeAngelis MM. Epidemiology of age-related macular degeneration (AMD): associations with cardiovascular disease phenotypes and lipid factors. Vol. 3, Eye and Vision. BioMed Central Ltd; 2016. \u003c/li\u003e\n\u003cli\u003eWong WL, Su X, Li X, Cheung CMG, Klein R, Cheng CY, et al. Global prevalence of age-related macular degeneration and disease burden projection for 2020 and 2040: A systematic review and meta-analysis. Lancet Glob Health. 2014 Feb;2(2). \u003c/li\u003e\n\u003cli\u003eSpaide RF, Jaffe GJ, Sarraf D, Freund KB, Sadda SR, Staurenghi G, et al. Consensus Nomenclature for Reporting Neovascular Age-Related Macular Degeneration Data: Consensus on Neovascular Age-Related Macular Degeneration Nomenclature Study Group. Ophthalmology. 2020 May 1;127(5):616\u0026ndash;36. \u003c/li\u003e\n\u003cli\u003eSchmidt-Erfurth U, Waldstein SM. A paradigm shift in imaging biomarkers in neovascular age-related macular degeneration. Vol. 50, Progress in Retinal and Eye Research. Elsevier Ltd; 2016. p. 1\u0026ndash;24. \u003c/li\u003e\n\u003cli\u003eSchmidt-Erfurth U, Vogl WD, Jampol LM, Bogunović H. Application of Automated Quantification of Fluid Volumes to Anti\u0026ndash;VEGF Therapy of Neovascular Age-Related Macular Degeneration. Ophthalmology. 2020 Sep 1;127(9):1211\u0026ndash;9. \u003c/li\u003e\n\u003cli\u003eBerta A, Larsen M, Macfadden W. Treat-and-Extend versus Monthly Regimen in Neovascular Age-Related Macular Degeneration Results with Ranibizumab from the TREND Study. 2017;57\u0026ndash;65. \u003c/li\u003e\n\u003cli\u003eBusbee BG, Ho AC, Brown DM, Heier JS, Su\u0026ntilde;er IJ. Twelve-Month Efficacy and Safety of 0 . 5 mg or 2 . 0 mg Ranibizumab in Patients with Subfoveal Neovascular Age-related Macular. 2013;1046\u0026ndash;56. \u003c/li\u003e\n\u003cli\u003eDugel PU, Koh A, Ogura Y, Jaffe GJ, Schmidt-Erfurth U, Brown DM, et al. HAWK and HARRIER: Phase 3, Multicenter, Randomized, Double-Masked Trials of Brolucizumab for Neovascular Age-Related Macular Degeneration. In: Ophthalmology. Elsevier Inc.; 2020. p. 72\u0026ndash;84. \u003c/li\u003e\n\u003cli\u003ePawloff M, Bogunovic H, Gruber A, Michl M, Riedl S, Schmidt-Erfurth U. SYSTEMATIC CORRELATION OF CENTRAL SUBFIELD THICKNESS WITH RETINAL FLUID VOLUMES QUANTIFIED BY DEEP LEARNING IN THE MAJOR EXUDATIVE MACULAR DISEASES [Internet]. Available from: http://links.lww.com/IAE/\u003c/li\u003e\n\u003cli\u003ePawloff M, Gerendas BS, Deak G, Bogunovic H, Gruber A, Schmidt-Erfurth U. Performance of retinal fluid monitoring in OCT imaging by automated deep learning versus human expert grading in neovascular AMD. Eye (Basingstoke). 2023;(December 2022):1\u0026ndash;8. \u003c/li\u003e\n\u003cli\u003eSchlegl T, Waldstein SM, Bogunovic H, Endstra\u0026szlig;er F, Sadeghipour A, Philip AM, et al. Fully Automated Detection and Quantification of Macular Fluid in OCT Using Deep Learning. Ophthalmology [Internet]. 2018;125(4):549\u0026ndash;58. Available from: https://doi.org/10.1016/j.ophtha.2017.10.031\u003c/li\u003e\n\u003cli\u003eWaldstein SM, Gerendas BS, Montuoro A, Simader C, Schmidt-Erfurth U. Quantitative comparison of macular segmentation performance using identical retinal regions across multiple spectral-domain optical coherence tomography instruments. British Journal of Ophthalmology. 2015 Jun 1;99(6):794\u0026ndash;800. \u003c/li\u003e\n\u003cli\u003eKlimscha S, Waldstein SM, Schlegl T, Bogunović H, Sadeghipour A, Philip AM, et al. Spatial correspondence between intraretinal fluid, subretinal fluid, and pigment epithelial detachment in neovascular age-related macular degeneration. Invest Ophthalmol Vis Sci. 2017 Aug 1;58(10):4039\u0026ndash;48. \u003c/li\u003e\n\u003cli\u003eSchmidt-Erfurth U, Reiter GS, Riedl S, Seeb\u0026ouml;ck P, Vogl WD, Blodi BA, et al. AI-based monitoring of retinal fluid in disease activity and under therapy. Vol. 86, Progress in Retinal and Eye Research. Elsevier Ltd; 2022. \u003c/li\u003e\n\u003cli\u003eBerta A, Larsen M, Macfadden W. Treat-and-Extend versus Monthly Regimen in Neovascular Age-Related Macular Degeneration Results with Ranibizumab from the TREND Study. 2017;57\u0026ndash;65. \u003c/li\u003e\n\u003cli\u003eCohen SY, Creuzot-Garcher C, Darmon J, Desmettre T, Korobelnik JF, Levrat F, et al. Types of choroidal neovascularisation in newly diagnosed exudative age-related macular degeneration. British Journal of Ophthalmology. 2007;91(9):1173\u0026ndash;6. \u003c/li\u003e\n\u003cli\u003eOlsen TW, Feng X, Kasper TJ, Rath PP, Steuer ER, Kertes PJ. Fluorescein angiographic lesion type frequency in neovascular age-related macular degeneration. Vol. 5, Evidence-Based Eye Care. 2004. p. 167\u0026ndash;9. \u003c/li\u003e\n\u003cli\u003eHaj Najeeb B, Deak G, Schmidt-erfurth U, Gerendas BS. THE RAP STUDY, REPORT TWO The Regional Distribution of Macular Neovascularization Type 3, a Novel Insight Into Its Etiology [Internet]. Available from: http://journals.lww.com/retinajournal\u003c/li\u003e\n\u003cli\u003eMettu PS, Allingham MJ, Cousins SW. Incomplete response to Anti-VEGF therapy in neovascular AMD: Exploring disease mechanisms and therapeutic opportunities. Vol. 82, Progress in Retinal and Eye Research. Elsevier Ltd; 2021. \u003c/li\u003e\n\u003cli\u003eHeier JS, Lad EM, Holz FG, Rosenfeld PJ, Guymer RH, Boyer D, et al. Articles Pegcetacoplan for the treatment of geographic atrophy secondary to age-related macular degeneration (OAKS and DERBY): two multicentre, randomised, double-masked, sham-controlled, phase 3 trials [Internet]. Vol. 402, www.thelancet.com. 2023. Available from: www.thelancet.com\u003c/li\u003e\n\u003cli\u003eKhanani AM, Patel SS, Staurenghi G, Tadayoni R, Danzig CJ, Eichenbaum DA, et al. Efficacy and safety of avacincaptad pegol in patients with geographic atrophy (GATHER2): 12-month results from a randomised, double-masked, phase 3 trial. The Lancet. 2023 Oct; \u003c/li\u003e\n\u003cli\u003eMichl M, Fabianska M, Seeb\u0026ouml;ck P, Sadeghipour A, Haj Najeeb B, Bogunovic H, et al. Automated quantification of macular fluid in retinal diseases and their response to anti-VEGF therapy. British Journal of Ophthalmology. 2022;106(1):113\u0026ndash;20. \u003c/li\u003e\n\u003cli\u003eRosenfeld PJ. Optical coherence tomography and the development of antiangiogenic therapies in neovascular age-related macular degeneration. Invest Ophthalmol Vis Sci. 2016;57(9):OCT14\u0026ndash;26. \u003c/li\u003e\n\u003cli\u003eChakravarthy U, Havilio M, Syntosi A, Pillai N, Wilkes E, Benyamini G, et al. Impact of macular fluid volume fluctuations on visual acuity during anti-VEGF therapy in eyes with nAMD. Eye (Basingstoke). 2021 Nov 1;35(11):2983\u0026ndash;90. \u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"age-related macular degeneration, artificial intelligence, deep learning, fluid quantification, optical coherence tomography, intraretinal fluid, subretinal fluid, pigment epithelial detachment","lastPublishedDoi":"10.21203/rs.3.rs-5313889/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5313889/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eTo investigate the association of manually measured retinal fluid by a human expert with AI-quantified retinal fluid volumes and to explore disease activity indicated by fluid volume distributions in neovascular age-related macular degeneration (nAMD) using an approved AI-based algorithm. This is a retrospective study analyzing baseline OCT data of patients with nAMD from multicenter study data. Manually measured maximum macular fluid heights vertically on B-scans in the central millimeter (CMM) for intraretinal fluid (IRF), subretinal fluid (SRF), and pigment epithelial detachment (PED) acquired from expert were associated with vertical fluid heights and three-dimensional volumes obtained by automated quantification using an AI-based tool (RetInSight Fluid Monitor Version 2). Out of 890 patients/eyes, we identified IRF in the CMM both manually and automatically in 328 eyes, SRF in 502 eyes, and PED in 705 eyes. The correlation between manual height and AI-based height was strong for IRF (r\u0026thinsp;=\u0026thinsp;0.87) and PED (r\u0026thinsp;=\u0026thinsp;0.91), and moderate for SRF (r\u0026thinsp;=\u0026thinsp;0.67). Manual height vs. AI-based volume correlation in the CMM was strong for IRF (r\u0026thinsp;=\u0026thinsp;0.76), and PED (r\u0026thinsp;=\u0026thinsp;0.87) and moderate for SRF (r\u0026thinsp;=\u0026thinsp;0.55). The correlation worsened when associating total fluid volumes in the central 6mm with manual CMM fluid height, indicating that CMM height does not represent total nAMD disease activity. AI-based fluid segmentation, in contrast to conventional human expert fluid measurements, provides a more comprehensive assessment, allowing for a significantly more accurate interpretation of total nAMD disease activity.\u003c/p\u003e","manuscriptTitle":"Correlation between human expert macular fluid height assessment and fluid volume quantification in neovascular age-related macular degeneration","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-11-11 05:59:21","doi":"10.21203/rs.3.rs-5313889/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-12-19T14:08:39+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-06-19T14:55:32+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"303491395092547940102267767941079005931","date":"2025-06-14T13:37:01+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"282901496658156449104975248860685856035","date":"2025-06-12T13:47:32+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-01-11T20:34:19+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"167199741623345336714557224103801960847","date":"2024-12-27T16:36:08+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-11-10T08:29:02+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-11-02T16:19:04+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2024-10-25T10:41:25+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-10-23T05:58:45+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2024-10-22T18:19:17+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"2a31b833-401b-4fc9-ae82-99dfc23f77ea","owner":[],"postedDate":"November 11th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":39776656,"name":"Health sciences/Diseases/Eye diseases/Macular degeneration"},{"id":39776657,"name":"Health sciences/Diseases/Eye diseases/Retinal diseases"}],"tags":[],"updatedAt":"2026-04-07T16:03:07+00:00","versionOfRecord":{"articleIdentity":"rs-5313889","link":"https://doi.org/10.1038/s41598-026-44982-8","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2026-03-31 15:59:43","publishedOnDateReadable":"March 31st, 2026"},"versionCreatedAt":"2024-11-11 05:59:21","video":"","vorDoi":"10.1038/s41598-026-44982-8","vorDoiUrl":"https://doi.org/10.1038/s41598-026-44982-8","workflowStages":[]},"version":"v1","identity":"rs-5313889","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5313889","identity":"rs-5313889","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}