Experienced Graders Research Articles

Association of subretinal drusenoid deposits and cardiovascular disease

The presence of subretinal drusenoid deposits (SDDs) in patients with age-related macular degeneration (AMD) appears to be correlated with cardiovascular disease (CVD) and may serve as a useful tool for predicting certain forms of CVD. Here, we tested this hypothesis by examining whether patients with AMD with SDDs are more likely to have a cardiovascular-related condition and/or undergo a cardiovascular procedure. This is a retrospective cohort study. We included a total of 597 patients with AMD either with SDDs (n = 332) or without SDDs (n = 265). All patients underwent spectral-domain optical coherence tomography (SD-OCT). The SD-OCT scans were annotated by masked, experienced graders. We also extracted data from the patients’ electronic medical records (EMRs), including patient demographics, cardiovascular diagnoses, and CVD-related procedures based on ICD-9 codes. AMD patients with SDDs were more likely to be diagnosed with CVD or undergo a cardiovascular procedure compared to AMD patients without SDDs, particularly percutaneous transluminal coronary angioplasty (PTCA; OR 2.73, 95% CI [1.21, 6.13], p = 0.02). Multivariate analysis confirmed the association between the presence of SDDs and a background of PTCA in the presence of other covariates. These data suggest that the presence of SDDs in patients with AMD correlates with certain severe cardiovascular conditions; SDDs and CVD may share common pathogenic pathways.

The effect of sampling window size on topographical maps of foveal cone density.

To characterize the effect of sampling window size on maps of foveal cone density derived from adaptive optics scanning light ophthalmoscope (AOSLO) images of the cone mosaic. Forty-four AOSLO-derived montages of the foveal cone mosaic (300 x 300µm) were used for this study (from 44 individuals with normal vision). Cone photoreceptor coordinates were semi-automatically identified by one experienced grader. From these coordinates, cone density matrices across each foveal montage were derived using 10 different sampling window sizes containing 5, 10, 15, 20, 40, 60, 80, 100, 150, or 200 cones. For all 440 density matrices, we extracted the location and value of peak cone density (PCD), the cone density centroid (CDC) location, and cone density at the CDC. Across all window sizes, PCD values were larger than those extracted at the CDC location, though the difference between these density values decreased as the sampling window size increased (p<0.0001). Overall, both PCD (r=-0.8099, p=0.0045) and density at the CDC (r=-0.7596, p=0.0108) decreased with increasing sampling window size. This reduction was more pronounced for PCD, with a 27.8% lower PCD value on average when using the 200-cone versus the 5-cone window (compared to only a 3.5% reduction for density at the CDC between these same window sizes). While the PCD and CDC locations did not occur at the same location within a given montage, there was no significant relationship between this PCD-CDC offset and sampling window size (p=0.8919). The CDC location was less variable across sampling windows, with an average per-participant 95% confidence ellipse area across the 10 window sizes of 47.56µm² (compared to 844.10µm² for the PCD location, p<0.0001). CDC metrics appear more stable across varying sampling window sizes than PCD metrics. Understanding how density values change according to the method used to sample the cone mosaic may facilitate comparing cone density data across different studies.

Validation of a deep learning model for automatic detection and quantification of five OCT critical retinal features associated with neovascular age-related macular degeneration

PurposeTo develop and validate a deep learning model for the segmentation of five retinal biomarkers associated with neovascular age-related macular degeneration (nAMD).Methods300 optical coherence tomography volumes from subject eyes with...

Self-Supervised Learning for Improved Optical Coherence Tomography Detection of Macular Telangiectasia Type 2

Deep learning image analysis often depends on large, labeled datasets, which are difficult to obtain for rare diseases. To develop a self-supervised approach for automated classification of macular telangiectasia type 2 (MacTel) on optical coherence tomography (OCT) with limited labeled data. This was a retrospective comparative study. OCT images from May 2014 to May 2019 were collected by the Lowy Medical Research Institute, La Jolla, California, and the University of Washington, Seattle, from January 2016 to October 2022. Clinical diagnoses of patients with and without MacTel were confirmed by retina specialists. Data were analyzed from January to September 2023. Two convolutional neural networks were pretrained using the Bootstrap Your Own Latent algorithm on unlabeled training data and fine-tuned with labeled training data to predict MacTel (self-supervised method). ResNet18 and ResNet50 models were also trained using all labeled data (supervised method). The ground truth yes vs no MacTel diagnosis is determined by retinal specialists based on spectral-domain OCT. The models' predictions were compared against human graders using accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), area under precision recall curve (AUPRC), and area under the receiver operating characteristic curve (AUROC). Uniform manifold approximation and projection was performed for dimension reduction and GradCAM visualizations for supervised and self-supervised methods. A total of 2636 OCT scans from 780 patients with MacTel and 131 patients without MacTel were included from the MacTel Project (mean [SD] age, 60.8 [11.7] years; 63.8% female), and another 2564 from 1769 patients without MacTel from the University of Washington (mean [SD] age, 61.2 [18.1] years; 53.4% female). The self-supervised approach fine-tuned on 100% of the labeled training data with ResNet50 as the feature extractor performed the best, achieving an AUPRC of 0.971 (95% CI, 0.969-0.972), an AUROC of 0.970 (95% CI, 0.970-0.973), accuracy of 0.898%, sensitivity of 0.898, specificity of 0.949, PPV of 0.935, and NPV of 0.919. With only 419 OCT volumes (185 MacTel patients in 10% of labeled training dataset), the ResNet18 self-supervised model achieved comparable performance, with an AUPRC of 0.958 (95% CI, 0.957-0.960), an AUROC of 0.966 (95% CI, 0.964-0.967), and accuracy, sensitivity, specificity, PPV, and NPV of 90.2%, 0.884, 0.916, 0.896, and 0.906, respectively. The self-supervised models showed better agreement with the more experienced human expert graders. The findings suggest that self-supervised learning may improve the accuracy of automated MacTel vs non-MacTel binary classification on OCT with limited labeled training data, and these approaches may be applicable to other rare diseases, although further research is warranted.

Segmentation of the fibrillar layer in advanced fuchs endothelial corneal dystrophy: Inter‐oberserver‐reliability and intra‐observer‐repeatability analysis of corneal densitometry images

Aims: Our previous studies state that a central subendothelial collagen‐rich fibrillar layer (FL) is present in about 80% of advanced Fuchs endothelial corneal dystrophy (FECD) corneas which correlates with decreased corneal endothelial cell density and focal oedema formation. In this study, we investigated the previously unknown inter‐observer reliability and intra‐observer reproducibility of FL segmentation on Scheimpflug corneal densitometry images in patients with advanced FECD.Methods: Preoperative Scheimpflug images of patients receiving DMEK or triple DMEK for advanced FECD were retrospectively analysed. FL‐negative cases were excluded. Images were manually segmented with the open‐source software MITK by two experienced graders. A MATLAB‐based code calculated the region of interest (ROI) area, maximum calliper length, maximum horizontal and vertical dimension, and Dice similarity coefficient for FL‐areas. The intraclass correlation coefficient (ICC) was calculated by IBM SPSS (two‐way mixed, average measures). The FL localization on corneal densitometry were visualized by spatial heatmaps by MATLAB.Results: Among 308 eyes, 248 were FL‐positive (80.5%) and 60 FL‐negative (19.5%). The average Dice similarity coefficient of FL areas segmented by two graders was calculated at 0.85. Calculations on the segmentations of Grader 1 vs. Grader 2 include: ROI 7.91mm2 ± 4.8 vs. 9.29 mm2 ± 5.0, Calliper length 4.07 mm ± 1.1 vs. 4.21 mm ± 1.0, maximum horizontal dimension 3.85 mm ± 1.1 vs. 3.95 mm ± 1.0, maximum vertical dimension 3.17 mm ± 1.0 vs. 3.3 mm ± 1.0. The ICC was calculated at 0.938 vs. 0.957 vs. 0.930 for ROI (0.955 vs. 0.960 vs. 0.944 calliper length, 0.958 vs. 0.959 vs. 0.946 maximum horizontal dimension, 0.952 vs. 0.941 vs. 0.945 maximum vertical dimension) and 0.85 vs. 0.86 vs. 0.89 for the Dice similarity coefficient (inter‐observer vs. intra‐observer Grader 1 vs. intra‐observer Grader 2). Spatial heatmap analysis demonstrated primarily inferotemporal localization of increased densitometry areas in FL‐positive eyes.Conclusions: FL segmentation on corneal densitometry images exhibit high inter‐observer reliability and intra‐observer reproducibility scores. We provide evidence for corneal densitometry images as a valid tool to detect FL in advanced FECD and confirm the inferotemporal corneal projection of FL.

Retro-Mode in NIDEK Mirante: A Comparative Analysis with Other Imaging Modalities for AMD and CSR.

Retro-mode is a novel technique capable of creating pseudo-3D images of the retina. However, its clinical utility remains unknown. This study aimed to evaluate the Nidek Mirante multimodal imaging platform for ocular assessment in patients with various retinal conditions. A total of 115 participants with central serous chorioretinopathy (CSR) and age-related macular degeneration (AMD) were included. Two experienced graders independently evaluated the images, and statistical analysis was performed to assess interclass correlation coefficients (ICC) between graders and modalities; Results: For CSR detection, retro-mode demonstrated exceptionally high ICC rates (ICC = 1; 100%), while color and autofluorescence (FAF) showed moderate coefficients (0.69 and 0.78, respectively). The detection of pigment epithelial detachment was high across all methods, with only retro-mode deviated right (DR) allowing detection in 69% of cases, while retro-mode DR and deviated left (DL) achieved 100% detection. FAF-green achieved a 95% detection rate. In detecting retinal atrophy, most modalities demonstrated high detection rates, with the lowest detection rates offered by retro-mode DL (ICC = 0.85) and DR (ICC = 0.89), while retro-mode ring aperture offered 0.97. Infra-red and fluorescein angiography imaging offered the highest detection rates among the tested modalities, with 97% and 100%, respectively. Retro-mode showed promise for comprehensive ocular evaluation and diagnosis, with certain imaging modalities demonstrating higher accuracy in detecting specific retinal features.

Automated Identification of Incomplete and Complete Retinal Epithelial Pigment and Outer Retinal Atrophy Using Machine Learning

To assess and validate a deep learning algorithm to automatically detect incomplete retinal pigment epithelial and outer retinal atrophy (iRORA) and complete retinal pigment epithelial and outer retinal atrophy (cRORA) in eyes with age-related macular degeneration. In a retrospective machine learning analysis, a deep learning model was trained to jointly classify the presence of iRORA and cRORA within a given B-scan. The algorithm was evaluated using 2 separate and independent datasets. OCT B-scan volumes from 71 patients with nonneovascular age-related macular degeneration captured at the Doheny-University of California Los Angeles Eye Centers and the following 2 external OCT B-scans testing datasets: (1) University of Pennsylvania, University of Miami, and Case Western Reserve University and (2) Doheny Image Reading Research Laboratory. The images were annotated by an experienced grader for the presence of iRORA and cRORA. A Resnet18 model was trained to classify these annotations for each B-scan using OCT volumes collected at the Doheny-University of California Los Angeles Eye Centers. The model was applied to 2 testing datasets to assess out-of-sample model performance. Model performance was quantified in terms of area under the receiver operating characteristic curve (AUROC) and area under the precision-recall curve (AUPRC). Sensitivity, specificity, and positive predictive value were also compared against additional clinician annotators. On an independently collected test set, consisting of 1117 volumes from the general population, the model predicted iRORA and cRORA presence within the entire volume with nearly perfect AUROC performance and AUPRC scores (iRORA, 0.61; 95% confidence interval [CI] [0.45, 0.82]: cRORA, 0.83; 95% CI [0.68, 0.95]). On another independently collected set, consisting of 60 OCT B-scans enriched for iRORA and cRORA lesions, the model performed with AUROC (iRORA: 0.68, 95% CI [0.54, 0.81]; cRORA: 0.84, 95% CI [0.75, 0.94]) and AUPRC (iRORA: 0.70, 95% CI [0.55, 0.86]; cRORA: 0.82, 95% CI [0.70, 0.93]). A deep learning model can accurately and precisely identify both iRORA and cRORA lesions within the OCT B-scan volume. The model can achieve similar sensitivity compared with human graders, which potentially obviates a laborious and time-consuming annotation process and could be developed into a diagnostic screening tool.

Tamoxifen Retinopathy and Macular Telangiectasia Type 2: Similarities and Differences on Multimodal Retinal Imaging

Tamoxifen-induced retinopathy (TR) and macular telangiectasia type 2 (MacTel) share a highly similar retinal phenotype. In this study, we aimed to evaluate differences and similarities that may point toward underlying mechanisms linking both disease entities. Retrospective, cross sectional study. Patients diagnosed with MacTel or TR. Patients underwent multimodal retinal imaging, including color fundus photography, spectral-domain OCT, fundus autofluorescence, fluorescein angiography, and OCT angiography (if available). Age, age of onset, best-corrected visual acuity, and bilaterality of changes were evaluated. Patients' eyes were graded for different morphologic characteristics by 4 experienced graders. Phenotypical characterization and comparison of frequencies of retinal characteristics of TR and MacTel on multimodal imaging. Twenty-eight eyes of 14 patients with TR and 118 eyes of 59 patients with MacTel were included. Age, age of onset, and best-corrected visual acuity were similar in both cohorts. All but 1 patient showed bilateral changes. In patients with MacTel, neurodegenerative changes and vascular alterations were equally present, whereas in patients with TR, neurodegenerative changes usually prevailed. Predilection sites within the central retina differed between the 2 diseases: most findings in patients with TR were limited to the foveal center, whereas changes in patients with MacTel were present throughout a slightly larger region ("MacTel area"), with an epicenter temporal to the foveal center. Distinct morphologic features included the distribution of retinal crystals, the size and position of ellipsoid zone breaks, and the presence of hyperreflective changes on OCT images. Focal hyperpigmentation and neovascular membranes were only present in eyes with MacTel. Macular telangiectasia and TR share a highly similar retinal phenotype, especially in early disease stages. Subtle differences on multimodal retinal images may help distinguish between these 2 disease entities. Our findings indicate the involvement of Müller cells in both diseases, which may explain the observed phenotypic characteristics and similarities.

Intraretinal Cysts as a Manifestation of Retinal Angiomatous Proliferation in Optical Coherence Tomography Angiography.

Background and Objectives: Intraretinal cysts are common pathology observed inspectral domain optical coherence tomography (SDOCT) in patients with neovascular form of age-related macular degeneration (AMD). The aim of the study was to determine if the presence of intraretinal cysts is positively correlated with diagnosis of retinal angiomatous proliferation (RAP) in optical coherence tomography angiography (OCTA). Material and Methods: A total of 21 eyes with intraretinal cysts in SDOCT exam (Group1) and 21 eyes with subretinal fluid(Group 2) were enrolled into the study. In each eye, the presence of intraretinal neovascularization (IRN) and chorioretinal anastomosis (CRA) was evaluated in OCTA by two experienced graders. Results: IRN was observed in 20 eyes (95.2%) from Group 1 and 5 eyes (23.8%) from Group 2. Features of CRA were found in 18 eyes (80.95%) and 16 eyes (76.2%) respectively for Group 1 and 2. Patients with cysts are 50 (95% CI: 5.43–460.52) times more likely to have IRN (p < 0.001). Conclusions: The presence of intraretinal cysts on SDOCT retinal sections in eyes with neovascular AMD corresponds to the presence of IRN on OCTA examination. The results indicate that the absence of a cyst does not exclude the presence of IRN and CRA which can be identified on OCTA.

Intraretinal Layer Segmentation Using Cascaded Compressed U-Nets.

Reliable biomarkers quantifying neurodegeneration and neuroinflammation in central nervous system disorders such as Multiple Sclerosis, Alzheimer’s dementia or Parkinson’s disease are an unmet clinical need. Intraretinal layer thicknesses on macular optical coherence tomography (OCT) images are promising noninvasive biomarkers querying neuroretinal structures with near cellular resolution. However, changes are typically subtle, while tissue gradients can be weak, making intraretinal segmentation a challenging task. A robust and efficient method that requires no or minimal manual correction is an unmet need to foster reliable and reproducible research as well as clinical application. Here, we propose and validate a cascaded two-stage network for intraretinal layer segmentation, with both networks being compressed versions of U-Net (CCU-INSEG). The first network is responsible for retinal tissue segmentation from OCT B-scans. The second network segments eight intraretinal layers with high fidelity. At the post-processing stage, we introduce Laplacian-based outlier detection with layer surface hole filling by adaptive non-linear interpolation. Additionally, we propose a weighted version of focal loss to minimize the foreground–background pixel imbalance in the training data. We train our method using 17,458 B-scans from patients with autoimmune optic neuropathies, i.e., multiple sclerosis, and healthy controls. Voxel-wise comparison against manual segmentation produces a mean absolute error of 2.3 μm, outperforming current state-of-the-art methods on the same data set. Voxel-wise comparison against external glaucoma data leads to a mean absolute error of 2.6 μm when using the same gold standard segmentation approach, and 3.7 μm mean absolute error in an externally segmented data set. In scans from patients with severe optic atrophy, 3.5% of B-scan segmentation results were rejected by an experienced grader, whereas this was the case in 41.4% of B-scans segmented with a graph-based reference method. The validation results suggest that the proposed method can robustly segment macular scans from eyes with even severe neuroretinal changes.

Automated vs. human evaluation of corneal staining

Background and purposeCorneal fluorescein staining is one of the most important diagnostic tests in dry eye disease (DED). Nevertheless, the result of this examination is depending on the grader. So far, there is no method for an automated quantification of corneal staining commercially available. Aim of this study was to develop a software-assisted grading algorithm and to compare it with a group of human graders with variable clinical experience in patients with DED.MethodsFifty images of eyes stained with 2 µl of 2% fluorescein presenting different severity of superficial punctate keratopathy in patients with DED were taken under standardized conditions. An algorithm for detecting and counting superficial punctate keratitis was developed using ImageJ with a training dataset of 20 randomly picked images. Then, the test dataset of 30 images was analyzed (1) by the ImageJ algorithm and (2) by 22 graders, all ophthalmologists with different levels of experience. All graders evaluated the images using the Oxford grading scheme for corneal staining at baseline and after 6–8 weeks. Intrarater agreement was also evaluated by adding a mirrored version of all original images into the set of images during the 2nd grading.ResultsThe count of particles detected by the algorithm correlated significantly (n = 30; p < 0.01) with the estimated true Oxford grade (Sr = 0,91). Overall human graders showed only moderate intrarater agreement (K = 0,426), while software-assisted grading was always the same (K = 1,0). Little difference was found between specialists and non-specialists in terms of intrarater agreement (K = 0,436 specialists; K = 0,417 non-specialists). The highest interrater agreement was seen with 75,6% in the most experienced grader, a cornea specialist with 29 years of experience, and the lowest was seen in a resident with 25,6% who had only 2 years of experience.ConclusionThe variance in human grading of corneal staining - if only small - is likely to have only little impact on clinical management and thus seems to be acceptable. While human graders give results sufficient for clinical application, software-assisted grading of corneal staining ensures higher consistency and thus is preferrable for re-evaluating patients, e.g., in clinical trials.

A Semi-Automated Method for Measuring Biofilm Accumulation on the Teeth Using Quantitative Light-Induced Fluorescence in Dogs and Cats.

Oral health conditions (eg, plaque, calculus, gingivitis) cause morbidity and pain in companion animals. Thus, developing technologies that can ameliorate the accumulation of oral biofilm, a critical factor in the progression of these conditions, is vital. Quantitative light-induced fluorescence (QLF) is a method to quantify oral substrate accumulation, and therefore, it can assess biofilm attenuation of different products. New software has recently been developed that automates aspects of the procedure. However, few QLF studies in companion animals have been performed. QLF was used to collect digital images of oral substrate accumulation on the teeth of dogs and cats to demonstrate the ability of QLF to discriminate between foods known to differentially inhibit oral substrate accumulation. Images were taken as a function of time and diet. Software developed by the Cytometry Laboratory, Purdue University quantified biofilm coverage. Intra- and intergrader reproducibility was also assessed, as was a comparison of the results of the QLF software with those of an experienced grader using undisclosed coverage-only metrics similar to those used for the Logan and Boyce index. Quantification of oral substrate accumulation using QLF-derived images demonstrated the ability to distinguish between dental diets known to differentially inhibit oral biofilm accumulation. Little variance in intra- and intergrader reproducibility was observed, and the comparison between the experienced Logan and Boyce grader and the QLF software yielded a concordance correlation coefficient of 0.89 (95% CI = 0.84, 0.92). These results show that QLF is a useful tool that allows the semi-automated quantification of the accumulation of oral biofilm in companion animals.

Modular deep neural networks for automatic quality control of retinal optical coherence tomography scans

Retinal optical coherence tomography (OCT) with intraretinal layer segmentation is increasingly used not only in ophthalmology but also for neurological diseases such as multiple sclerosis (MS). Signal quality influences segmentation results, and high-quality OCT images are needed for accurate segmentation and quantification of subtle intraretinal layer changes. Among others, OCT image quality depends on the ability to focus, patient compliance and operator skills. Current criteria for OCT quality define acceptable image quality, but depend on manual rating by experienced graders and are time consuming and subjective. In this paper, we propose and validate a standardized, grader-independent, real-time feedback system for automatic quality assessment of retinal OCT images. We defined image quality criteria for scan centering, signal quality and image completeness based on published quality criteria and typical artifacts identified by experienced graders when inspecting OCT images. We then trained modular neural networks on OCT data with manual quality grading to analyze image quality features. Quality analysis by a combination of these trained networks generates a comprehensive quality report containing quantitative results. We validated the approach against quality assessment according to the OSCAR-IB criteria by an experienced grader. Here, 100 OCT files with volume, circular and radial scans, centered on optic nerve head and macula, were analyzed and classified. A specificity of 0.96, a sensitivity of 0.97 and an accuracy of 0.97 as well as a Matthews correlation coefficient of 0.93 indicate a high rate of correct classification. Our method shows promising results in comparison to manual OCT grading and may be useful for real-time image quality analysis or analysis of large data sets, supporting standardized application of image quality criteria.

Rotational Stability of Intraocular Lenses: A Standardized Method for More Accurate Measurements in Future Studies

To assess the reliability and reproducibility of a new semiautomated evaluation method, "Rotix," for intraocular lens (IOL) rotation and to define a standardized evaluation method for future toric IOL studies. Reliability and reproducibility study. Setting: Department of Ophthalmology, Medical University of Vienna. A dataset of 25 patients with 2 consecutive follow-up visits was created to test the intra- and interrater reliability. A data test set of 10 patients including 30 pictures taken 5 minutes apart was created to test the short-term reproducibility. Evaluation of IOL rotational stability using nontoric implants in 25×2 consecutive follow-up visits. Two experienced graders performed axis evaluation 3 times in a randomized order. One experienced grader performed axis evaluation for the short-term reproducibility study. Reference landmarks at the end of operation were used to assess the IOL axis. Intra- and interrater reliability and short-term reproducibility of axis measurements. Mean standard deviation for intrarater reliability was 0.16 degrees. The intraclass correlation coefficients were 0.97 for grader 1 and 0.96 for grader 2. A very high interrater correlation of 0.95 was found. The mean individual difference between grader 1 and grader 2 was 0.061 ± 0.28 degrees. Short-term reproducibility showed a mean standard deviation of 0.22 ± 0.14 degrees. The novel semiautomated evaluation method showed an accurate inter- and intrarater reliability. Short-term reproducibility was below 0.25 degrees. The method of using nonmovable reference landmarks showed reliable results and should be used as a standard in future toric IOL studies.

Development of a standardized histopathology scoring system for intervertebral disc degeneration in rat models: An initiative of the ORS spine section.

BackgroundRats are a widely accepted preclinical model for evaluating intervertebral disc (IVD) degeneration and regeneration. IVD morphology is commonly assessed using histology, which forms the foundation for quantifying the state of IVD degeneration. IVD degeneration severity is evaluated using different grading systems that focus on distinct degenerative features. A standard grading system would facilitate more accurate comparison across laboratories and more robust comparisons of different models and interventions.AimsThis study aimed to develop a histology grading system to quantify IVD degeneration for different rat models.Materials & MethodsThis study involved a literature review, a survey of experts in the field, and a validation study using 25 slides that were scored by 15 graders from different international institutes to determine inter‐ and intra‐rater reliability.ResultsA new IVD degeneration grading system was established and it consists of eight significant degenerative features, including nucleus pulposus (NP) shape, NP area, NP cell number, NP cell morphology, annulus fibrosus (AF) lamellar organization, AF tears/fissures/disruptions, NP‐AF border appearance, as well as endplate disruptions/microfractures and osteophyte/ossification. The validation study indicated this system was easily adopted, and able to discern different severities of degenerative changes from different rat IVD degeneration models with high reproducibility for both experienced and inexperienced graders. In addition, a widely‐accepted protocol for histological preparation of rat IVD samples based on the survey findings include paraffin embedding, sagittal orientation, section thickness < 10 μm, and staining using H&E and/or SO/FG to facilitate comparison across laboratories.ConclusionThe proposed histological preparation protocol and grading system provide a platform for more precise comparisons and more robust evaluation of rat IVD degeneration models and interventions across laboratories.

Quantification of Key Retinal Features in Early and Late Age-Related Macular Degeneration Using Deep Learning

We sought to develop and validate a deep learning model for segmentation of 13 features associated with neovascular and atrophic age-related macular degeneration (AMD). Development and validation of a deep-learning model for feature segmentation. Data for model development were obtained from 307 optical coherence tomography volumes. Eight experienced graders manually delineated all abnormalities in 2712 B-scans. A deep neural network was trained with these data to perform voxel-level segmentation of the 13 most common abnormalities (features). For evaluation, 112 B-scans from 112 patients with a diagnosis of neovascular AMD were annotated by 4 independent observers. The main outcome measures were Dice score, intraclass correlation coefficient, and free-response receiver operating characteristic curve. On 11 of 13 features, the model obtained a mean Dice score of 0.63 ± 0.15, compared with 0.61 ± 0.17 for the observers. The mean intraclass correlation coefficient for the model was 0.66 ± 0.22, compared with 0.62 ± 0.21 for the observers. Two features were not evaluated quantitatively because of a lack of data. Free-response receiver operating characteristic analysis demonstrated that the model scored similar or higher sensitivity per false positives compared with the observers. The quality of the automatic segmentation matches that of experienced graders for most features, exceeding human performance for some features. The quantified parameters provided by the model can be used in the current clinical routine and open possibilities for further research into treatment response outside clinical trials.

Cone Identification in Choroideremia: Repeatability, Reliability, and Automation Through Use of a Convolutional Neural Network.

PurposeAdaptive optics imaging has enabled the visualization of photoreceptors both in health and disease. However, there remains a need for automated accurate cone photoreceptor identification in images of disease. Here, we apply an open-source convolutional neural network (CNN) to automatically identify cones in images of choroideremia (CHM). We further compare the results to the repeatability and reliability of manual cone identifications in CHM.MethodsWe used split-detection adaptive optics scanning laser ophthalmoscopy to image the inner segment cone mosaic of 17 patients with CHM. Cones were manually identified twice by one experienced grader and once by two additional experienced graders in 204 regions of interest (ROIs). An open-source CNN either pre-trained on normal images or trained on CHM images automatically identified cones in the ROIs. True and false positive rates and Dice's coefficient were used to determine the agreement in cone locations between data sets. Interclass correlation coefficient was used to assess agreement in bound cone density.ResultsIntra- and intergrader agreement for cone density is high in CHM. CNN performance increased when it was trained on CHM images in comparison to normal, but had lower agreement than manual grading.ConclusionsManual cone identifications and cone density measurements are repeatable and reliable for images of CHM. CNNs show promise for automated cone selections, although additional improvements are needed to equal the accuracy of manual measurements.Translational RelevanceThese results are important for designing and interpreting longitudinal studies of cone mosaic metrics in disease progression or treatment intervention in CHM.

Convolutional Neural Networks for Automatic Risser Stage Assessment.

To develop an automatic method for the assessment of the Risser stage using deep learning that could be used in the management panel of adolescent idiopathic scoliosis (AIS). In this institutional review board approved-study, a total of 1830 posteroanterior radiographs of patients with AIS (age range, 10-18 years, 70% female) were collected retrospectively and graded manually by six trained readers using the United States Risser staging system. Each radiograph was preprocessed and cropped to include the entire pelvic region. A convolutional neural network was trained to automatically grade conventional radiographs according to the Risser classification. The network was then validated by comparing its accuracy against the interobserver variability of six trained graders from the authors' institution using the Fleiss κ statistical measure. Overall agreement between the six observers was fair, with a κ coefficient of 0.65 for the experienced graders and agreement of 74.5%. The automatic grading method obtained a κ coefficient of 0.72, which is a substantial agreement with the ground truth, and an overall accuracy of 78.0%. The high accuracy of the model presented here compared with human readers suggests that this work may provide a new method for standardization of Risser grading. The model could assist physicians with the task, as well as provide additional insights in the assessment of bone maturity based on radiographs.© RSNA, 2020.

Optimising subjective anterior eye grading precision

PurposeTo establish the optimum grading increment which ensured parity between practitioners while maximising clinical precision. MethodsSecond year optometry students (n = 127, 19.5 ± 1.4 years, 55 % female) and qualified eye care practitioners (n = 61, 40.2 ± 14.8 years, 52 % female) had 30 s to grade each of bulbar, limbal and palpebral hyperaemia of the upper lid of 4 patients imaged live with a digital slit lamp under 16× magnification, diffuse illumination, with the image projected on a screen. The patients were presented in a randomised sequence 3 times in succession, during which the graders used the Efron printed grading scale once to the nearest 0.1 increment, once to nearest 0.5 increment and once to the nearest integer grade in a randomised order. Graders were masked to their previous responses. ResultsFor most grading conditions less than 20 % of clinicians showed a ≤0.1 difference in grade from the mean. In contrast, more than 50 % of the student graders and 40 % of experienced graders showed a difference in grade from the mean within 0.5 for all conditions under measurement. Student precision in grading was better with both 0.1 and 0.5 grading increments than grading to the nearest unit, except for limbal hyperaemia where they performed more accurately with 0.5 unit increment grading. Limbal grading precision was not affected by grading step increment for experienced practitioners, but 0.1 and 0.5 grading increments were both better than the 1.0 grading increment for bulbar hyperaemia and the 0.1 grading increment was better than the 0.5 grading increment and both were better than the 1.0 grading increment for palpebral hyperaemia. ConclusionAlthough narrower interval scales maximise the ability to detect smaller clinical changes, the grading increment should not exceed one standard deviation of the discrepancy between measurements. Therefore, 0.5 grading increments are recommended for subjective anterior eye physiology grading (limbal, bulbar and palpebral redness).

A Deep Learning Model for Segmentation of Geographic Atrophy to Study Its Long-Term Natural History

To develop and validate a deep learning model for the automatic segmentation of geographic atrophy (GA) using color fundus images (CFIs) and its application to study the growth rate of GA. Prospective, multicenter, natural history study with up to 15 years of follow-up. Four hundred nine CFIs of 238 eyes with GA from the Rotterdam Study (RS) and Blue Mountain Eye Study (BMES) for model development, and 3589 CFIs of 376 eyes from the Age-Related Eye Disease Study (AREDS) for analysis of GA growth rate. A deep learning model based on an ensemble of encoder-decoder architectures was implemented and optimized for the segmentation of GA in CFIs. Four experienced graders delineated, in consensus, GA in CFIs from the RS and BMES. These manual delineations were used to evaluate the segmentation model using 5-fold cross-validation. The model was applied further to CFIs from the AREDS to study the growth rate of GA. Linear regression analysis was used to study associations between structural biomarkers at baseline and the GA growth rate. A general estimate of the progression of GA area over time was made by combining growth rates of all eyes with GA from the AREDS set. Automatically segmented GA and GA growth rate. The model obtained an average Dice coefficient of 0.72±0.26 on the BMES and RS set while comparing the automatically segmented GA area with the graders' manual delineations. An intraclass correlation coefficient of 0.83 was reached between the automatically estimated GA area and the graders' consensus measures. Nine automatically calculated structural biomarkers (area, filled area, convex area, convex solidity, eccentricity, roundness, foveal involvement, perimeter, and circularity) were significantly associated with growth rate. Combining all growth rates indicated that GA area grows quadratically up to an area of approximately 12 mm2, after which growth rate stabilizes or decreases. The deep learning model allowed for fully automatic and robust segmentation of GA on CFIs. These segmentations can be used to extract structural characteristics of GA that predict its growth rate.