Management Of Anti-Vegf Injections For Neovascular Amd In A Surgical Clinic: Improving Automated Diagnosis And Management Of Retinal Comorbidities In The Cataract Population

Age-related macular degeneration (AMD) is one of the most common comorbidities in the elderly population and an important incidental finding during pre-surgical examinations. Real-life treatment outcomes in neovascular AMD (nAMD) are far from satisfactory compared to randomized clinical trials (RCT). In RCT retinal imaging is often assessed by certified readers in a central reading center. However, real-world interpretations of retinal images are lacking precision and repeatability. This study was conducted to report on the performance of the Vienna Fluid Monitor for the quantification of macular fluid (subretinal fluid (SRF), intraretinal fluid (IRF) and pigment epithelium detachment (PED)) in nAMD.

Optical coherence tomography (OCT) data from a real-world outpatient clinic in a tertiary referral center (Department of Ophthalmology at the Medical University of Vienna, Austria).

A multi-class deep learning-based method was developed to identify regions of SRF, IRF and PED in nAMD subjects. The dataset was comprised of 219 OCT scans from Heidelberg Spectralis HRA+OCT devices (Heidelberg Engineering, Heidelberg, Germany) from 155 different nAMD patients. The dataset included data from patients with type 1, 2 and 3 nAMD from both treatment-naïve and post-treatment stage. Data augmentation was used to address variable image quality, and several types of OCT specific artefacts found in real world data. The performance of the algorithm was evaluated on a pixel-level basis for IRF, SRF and PED against ground truth of manual reading made by a reading expert.

155 patients were split into three sets: 99 for training, 28 to validate model selection and 28 for the test set used to report performance. The validation and testing sets were selected randomly and supplemented by data with confounding morphological characteristics. There was no patient overlap between the separate sets. The Pearson correlations in the central 1mm for each quantitative fluid volume showed strong correlations between the algorithm and the expert gradings (IRF: r=0.9998, p<0.001; SRF: r=0.9940, p<0.001, and PED: r=0.9848, p<0.001). Fluid detection was consistently well performed by the algorithm for each compartment on the b-scan level. Sensitivity and precision on a pixel level were high for each fluid type.

Disease activity in nAMD can be measured in a precise and repeatable manner using AI-based decision support tools. Quantifications of macular fluids enable clinicians to base their re-treatment criteria on objective measurements that can be implemented in real-time in a busy surgical clinic. With the use of accurate AI-based fluid quantification every clinician has the possibility to access reading center-like precision in determination of disease activity. Going beyond the current practice will introduce true precision medicine and personalized treatment regimens into daily routine along with easing the burden on health-care providers, treating clinicians and patients.