Supervised learning and dimension reduction techniques for quantification of retinal fluid in optical coherence tomography images

Explainable ensemble learning method for OCT detection with transfer learning

The accuracy and interpretability of artificial intelligence (AI) are crucial for the advancement of optical coherence tomography (OCT) image detection, as it can greatly reduce the manual labor required by clinicians. By prioritizing these aspects during development and application, we can make significant progress towards streamlining the clinical workflow. In this paper, we propose an explainable ensemble approach that utilizes transfer learning to detect fundus lesion diseases through OCT imaging. Our study utilized a publicly available OCT dataset consisting of normal subjects, patients with dry age-related macular degeneration (AMD), and patients with diabetic macular edema (DME), each with 15 samples. The impact of pre-trained weights on the performance of individual networks was first compared, and then these networks were ensemble using majority soft polling. Finally, the features learned by the networks were visualized using Grad-CAM and CAM. The use of pre-trained ImageNet weights improved the performance from 68.17% to 92.89%. The ensemble model consisting of the three CNN models with pre-trained parameters loaded performed best, correctly distinguishing between AMD patients, DME patients and normal subjects 100% of the time. Visualization results showed that Grad-CAM could display the lesion area more accurately. It is demonstrated that the proposed approach could have good performance of both accuracy and interpretability in retinal OCT image detection.

Multi-Compartment Spatially-Derived Radiomics From Optical Coherence Tomography Predict Anti-VEGF Treatment Durability in Macular Edema Secondary to Retinal Vascular Disease: Preliminary Findings

OBJECTIVE

Diabetic macular edema (DME) and retinal vein occlusion (RVO) are the leading causes of visual impairments across the world. Vascular endothelial growth factor (VEGF) stimulates breakdown of blood-retinal barrier that causes accumulation of fluid within macula. Anti-VEGF therapy is the first-line treatment for both the diseases; however, the degree of response varies for individual patients. The main objective of this work was to identify the (i) texture-based radiomics features within individual fluid and retinal tissue compartments of baseline spectral-domain optical coherence tomography (SD-OCT) images and (ii) the specific spatial compartments that contribute most pertinent features for predicting therapeutic response.

METHODS

A total of 962 texture-based radiomics features were extracted from each of the fluid and retinal tissue compartments of OCT images, obtained from the PERMEATE study. Top-performing features selected from the consensus of different feature selection methods were evaluated in conjunction with four different machine learning classifiers: Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA), Random Forest (RF), and Support Vector Machine (SVM) in a cross-validated approach to distinguish eyes tolerating extended interval dosing (non-rebounders) and those requiring more frequent dosing (rebounders).

RESULTS

Combination of fluid and retinal tissue features yielded a cross-validated area under receiver operating characteristic curve (AUC) of 0.78±0.08 in distinguishing rebounders from non-rebounders.

CONCLUSIONS

This study revealed that the texture-based radiomics features pertaining to IRF subcompartment were most discriminating between rebounders and non-rebounders to anti-VEGF therapy. Clinical Impact: With further validation, OCT-based imaging biomarkers could be used for treatment management of DME patients.

Longitudinal Higher-Order OCT Assessment of Quantitative Fluid Dynamics and the Total Retinal Fluid Index in Neovascular AMD

Purpose

The purpose of this study was to evaluate the feasibility of assessing quantitative longitudinal fluid dynamics and total retinal fluid indices (TRFIs) with higher-order optical coherence tomography (OCT) for neovascular age-related macular degeneration (nAMD).

Methods

A post hoc image analysis study was performed using the phase II OSPREY clinical trial comparing brolucizumab and aflibercept in nAMD. Higher-order OCT analysis using a machine learning−enabled fluid feature extraction platform was used to segment intraretinal fluid (IRF) and subretinal fluid (SRF) volumetric components. TRFI, the proportion of fluid volume against total retinal volume, was calculated. Longitudinal fluid metrics were evaluated for the following groups: all subjects (i.e. treatment agnostic), brolucizumab, and aflibercept.

Results

Mean IRF and SRF volumes were significantly reduced from baseline at each timepoint for all groups. Fluid feature extraction allowed high-resolution assessment of quantitative fluid burden. A greater proportion of brolucizumab participants achieved true zero and minimal fluid (total fluid volume between 0.0001 and 0.001mm³) versus aflibercept participants at week 40. True zero fluid during q12 brolucizumab dosing was achieved in 36.6% to 38.5%, similar to the 25.6% to 38.5% during the corresponding q8 aflibercept cycles. TRFI was significantly reduced from baseline in all groups.

Conclusions

Higher-order OCT analysis demonstrates the feasibility of fluid feature extraction and longitudinal volumetric fluid burden and TRFI characterization in nAMD, supporting a unique opportunity for fluid burden assessment and the impact on outcomes.

Translational Relevance

Detection and characterization of disease activity is vital for optimal treatment of nAMD. Longitudinal assessment of fluid dynamics and the TRFI provide important proof of concept for future automated tools in characterizing disease activity.

[Artificial intelligence in management of macular edema : Opportunities and challenges]

Macular edema occurs in a wide variety of ophthalmological diseases. The diagnostics and treatment are an important part of modern ophthalmology. Due to the continuous development, artificial intelligence (AI) offers many opportunities to improve the management of macular edema. This article provides the readership with an overview of this interesting topic.

A Novel Approach to Quantitative Evaluation of Outer Retinal Lesions Via a New Parameter "Integral" in Spectral Domain Optical Coherence Tomography

Purpose

The purpose of this study was to design a new parameter "integral" to quantitatively evaluate the spatial cumulative reflectivity of the outer retinal layers in optical coherence tomography (OCT), and to investigate its role in the detection of outer retinal diseases.

Methods

This was a cross-sectional study. Fovea-centered line OCT scans were performed on 60 eyes of 60 healthy volunteers and 44 eyes of 44 patients diagnosed with outer retinal diseases. The integrals of the ellipsoid zone (EZ) and interdigitation zone (IZ) were measured by respectively accumulating the grayscale values of all the pixels within the EZ and IZ at specified locations on the scanning lines, and were then adjusted by calculating their percentages on the outer retina. The integrals of the EZ and IZ were compared between the two groups.

Results

The integrals of the EZ and IZ were stably and normally distributed in the healthy eyes, and were significantly lower in eyes with outer retinal lesions than in healthy ones (P < 0.05). Moreover, the integrals of the EZ and IZ were correlated with best corrected visual acuity (BCVA; adjusted R² = 0.620) and the presence of outer retinal lesions (Nagelkerke R² = 0.767). The area under the receiver operating characteristic (ROC) curve was 0.954 (95% confidence interval [CI] = 0.918-0.990) when the integral was selected as a diagnostic variable.

Conclusions

Obtained from this novel quantification method, the new parameter integral was comparable between different individuals and had the potential to detect outer retinal abnormalities in reflectivity through OCT.

Translational Relevance

Our work verified the feasibility of the new image analysis technique in the detection of the diseases affecting the outer retina.

Automatic detection of parapapillary atrophy and its association with children myopia

BACKGROUND AND OBJECTIVE

To develop an automatic parapapillary atrophy (PPA) detection algorithm in retinal fundus images and discuss the association between PPA and myopia to facilitate diagnosis and prediction of children myopia.

METHODS

The proposed algorithm consists of PPA identification and segmentation, which are evaluated by comparing with ophthalmologist's annotation. The association between PPA parameters and myopia is analyzed via Spearman correlation.

RESULTS

The accuracy of PPA identification reaches 90.78%. The F1-score of PPA segmentation is 0.67, and the Pearson correlation between the automatic measurement and ground truths for the area of PPA (A_PPA), the ratio (μ) of A_PPA to the area of optic disc (OD) and the maximal width of PPA (W) are 0.74, 0.60, and 0.69 (all p < 0.001). All these parameter changes are significantly correlated with the change of ratio of axial length to corneal curvature (ΔALCC), spherical equivalent (ΔSE), and axial length (ΔAL) (all p < 0.01), in which the highest association is 0.75 between ΔW (the change of W) and ΔALCC.

CONCLUSIONS

The proposed algorithm can provide accurate PPA measurement. Strong association between the changes of PPA and the progress of children myopia are observed and the width of PPA has the best association among three PPA parameters.

Higher-Order Assessment of OCT in Diabetic Macular Edema from the VISTA Study: Ellipsoid Zone Dynamics and the Retinal Fluid Index

PURPOSE

To investigate retinal fluid features and ellipsoid zone (EZ) integrity dynamics on spectral-domain OCT (SD-OCT) in eyes with diabetic macular edema (DME) treated with intravitreal aflibercept injection (IAI) in the VISTA-DME study.

DESIGN

A post hoc subanalysis of a phase III, prospective clinical trial.

PARTICIPANTS

Eyes received either IAI 2 mg every 4 weeks (2q4) or every 8 weeks after 5 initial monthly doses (2q8).

METHODS

All eyes from the VISTA Phase III study in the IAI groups imaged with the Cirrus HD-OCT system (Zeiss, Oberkochen, Germany) were included. The OCT macular cube datasets were evaluated using a novel software platform to generate retinal layer and fluid boundary lines that were manually corrected for assessment of change in EZ parameters and volumetric fluid parameters from baseline. The retinal fluid index (i.e., proportion of the retinal volume consisting of cystic fluid) was also calculated at each time point.

MAIN OUTCOME MEASURES

The feasibility of volumetric assessment of higher-order OCT-based retinal parameters and its correlation with best-corrected visual acuity (BCVA).

RESULTS

Overall, 106 eyes of 106 patients were included. Specifically, 52 eyes of 52 patients were included in the IAI 2q4 arm, and 54 eyes of 54 patients were included in the IAI 2q8 arm. Ellipsoid zone integrity metrics significantly improved from baseline to week 100, including central macular mean EZ to retinal pigment epithelium (RPE) thickness (2q4: 26.6 μm to 31.6 μm, P < 0.001; 2q8: 25.2 μm to 31.4 μm, P < 0.001). At week 100, central macular intraretinal fluid volume was reduced by >65% (P < 0.001) and central macular subretinal fluid volume was reduced by >99% in both arms (P < 0.001). Central macular retinal fluid index (RFI) significantly improved in both arms (2q4: 17.9% to 7.2%, P < 0.001; 2q8: 19.8% to 4.2%, P < 0.001). Central macular mean EZ-RPE thickness (i.e., a surrogate for photoreceptor outer segment length) and central RFI were independently correlated with BCVA at multiple follow-up visits.

CONCLUSIONS

Intravitreal aflibercept injection resulted in significant improvement in EZ integrity and quantitative fluid metrics in both 2q4 and 2q8 arms and correlated with visual function.

Deep learning based retinal OCT segmentation

We look at the recent application of deep learning (DL) methods in automated fine-grained segmentation of spectral domain optical coherence tomography (OCT) images of the retina. We describe a new method combining fully convolutional networks (FCN) with Gaussian Processes for post processing. We report performance comparisons between the proposed approach, human clinicians, and other machine learning (ML) such as graph based approaches. The approach is demonstrated on an OCT dataset consisting of mild non-proliferative diabetic retinopathy from the University of Miami. The method is shown to have performance on par with humans, also compares favorably with the other ML methods, and appears to have as small or smaller mean unsigned error (equal to 1.06), versus errors ranging from 1.17 to 1.81 for other methods, and compared with human error of 1.10.

Correlation between morphological characteristics in spectral-domain-optical coherence tomography, different functional tests and a patient's subjective handicap in acute central serous chorioretinopathy

PURPOSE

The purpose of this study was to identify quantitatively measurable morphologic optical coherence tomography (OCT) characteristics in patients with an acute episode of central serous chorioretinopathy (CSC) and evaluate their correlation to functional and psychological variables for their use in daily clinical practice.

METHODS

Retinal thickness (RT), the height, area and volume of subretinal fluid (SRF)/pigment epithelium detachments were evaluated using the standardized procedures of the Vienna Reading Center. These morphologic characteristics were compared with functional variables [best-corrected visual acuity (BCVA), contrast sensitivity (CS), retinal sensitivity/microperimetry, fixation stability], and patients' subjective handicap from CSC using the National Eye Institute 25-item Visual Function Questionnaire (NEI VFQ-25).

RESULTS

Data from 39 CSC patients were included in this analysis. Three different SRF height measures showed a high negative correlation (r = -0.7) to retinal sensitivity within the central 9°, which was also negatively correlated with SRF area and volume (r = -0.6). The CS score and fixation stability (fixation points within 2°) showed a moderate negative correlation (r = -0.4) with SRF height variables. Comparison of the subjective handicap with morphological characteristics in spectral-domain (SD)-OCT showed SRF height had the highest correlation (r = -0.4) with the subjective problems reported and overall NEI VFQ-25 score.

CONCLUSION

In conclusion, SRF height measured in SD-OCT showed the best correlation with functional variables and patients' subjective handicap caused by the disease and therefore seems to be the best variable to look at in daily clinical routine. Even though area and volume also show a correlation, these cannot be so easily measured as height and are therefore not suggested for daily clinical routine.

Artificial intelligence in retina

Major advances in diagnostic technologies are offering unprecedented insight into the condition of the retina and beyond ocular disease. Digital images providing millions of morphological datasets can fast and non-invasively be analyzed in a comprehensive manner using artificial intelligence (AI). Methods based on machine learning (ML) and particularly deep learning (DL) are able to identify, localize and quantify pathological features in almost every macular and retinal disease. Convolutional neural networks thereby mimic the path of the human brain for object recognition through learning of pathological features from training sets, supervised ML, or even extrapolation from patterns recognized independently, unsupervised ML. The methods of AI-based retinal analyses are diverse and differ widely in their applicability, interpretability and reliability in different datasets and diseases. Fully automated AI-based systems have recently been approved for screening of diabetic retinopathy (DR). The overall potential of ML/DL includes screening, diagnostic grading as well as guidance of therapy with automated detection of disease activity, recurrences, quantification of therapeutic effects and identification of relevant targets for novel therapeutic approaches. Prediction and prognostic conclusions further expand the potential benefit of AI in retina which will enable personalized health care as well as large scale management and will empower the ophthalmologist to provide high quality diagnosis/therapy and successfully deal with the complexity of 21st century ophthalmology.