Artificial intelligence in retina

Schmidt‐Erfurth, Ursula; Sadeghipour, Amir; Gerendas, Bianca S.; Waldstein, Sebastian M.; Bogunović, Hrvoje

doi:10.1016/j.preteyeres.2018.07.004

Cited by 585 publications

(411 citation statements)

References 155 publications

Supporting

Mentioning

399

Contrasting

Unclassified

Order By: Relevance

“…As the morphological and hemodynamic parameters of the bulbar conjunctival vessels could potentially be used to access the status of ocular surface diseases, large scale clinical studies will need to be conducted to comprehensively characterize the relations between these quantitative parameters of human bulbar conjunctival microvasculature and differences of ocular diseases. What is more, while artificial intelligence has been recognized to be so promising in ophthalmic disease diagnosis with different ophthalmic images [49][50][51], our multi-modal system can potentially offer a technical basis for providing complementary images for multimodal artificial intelligence applications.…”

Section: Discussionmentioning

confidence: 99%

Multi-modal Anterior Eye Imager Combining Ultra-High Resolution OCT and Microvascular Imaging for Structural and Functional Evaluation of the Human Eye

et al. 2020

View full text Add to dashboard Cite

To establish complementary information for the diagnosis and evaluation of ocular surface diseases, we developed a multi-modal, non-invasive optical imaging platform by combining ultra-high resolution optical coherence tomography (UHR-OCT) with a microvascular imaging system based on slit-lamp biomicroscopy. Our customized UHR-OCT module achieves an axial resolution of ≈2.9 µm in corneal tissue with a broadband light source and an A-line acquisition rate of 24 kHz with a line array CCD camera. The microvascular imaging module has a lateral resolution of 3.5 µm under maximum magnification of ≈187.5× with an imaging rate of 60 frames/s, which is sufficient to image the conjunctival vessel network and record the movement trajectory of clusters of red blood cells. By combining the imaging optical paths of different modules, our customized multi-modal anterior eye imaging platform is capable of performing real-time cross-sectional UHR-OCT imaging of the anterior eye, conjunctival vessel network imaging, high-resolution conjunctival blood flow videography, fluorescein staining and traditional slit-lamp imaging on a single device. With self-developed software, a conjunctival vessel network image and blood flow videography were further analyzed to acquire quantitative morphological and hemodynamics parameters, including vessel fractal dimensions, blood flow velocity and vessel diameters. The ability of our multi-modal anterior eye imager to provide both structural and functional information for ophthalmic clinical applications was demonstrated on a healthy human subject and a keratitis patient. evaluation of ocular surface diseases, different imaging modalities are usually applied in ophthalmic clinics to provide multi-modal and multi-dimensional images [7,8].Slit-lamp biomicroscopy (SLB) is the gold standard for routine ophthalmic examinations. By shining a thin sheet of light into the eye, an in vivo stereoscopic evaluation of the pathological changes of the cornea, conjunctiva and other anterior segment tissues can be performed by SLB [9]. However, the limited magnification and poor spatial resolution of traditional SLB make it difficult to visualize more detailed pathological structures and perform accurate biometric measurements [10][11][12]. Fluorescein stain could stain stroma while leaving the intact epithelium with a specific fluorescein dye, thereby demarcating the location and area of the epithelial loss in ocular surface diseases, on a relative macro scale as well [13]. Developed in the 1990s, optical coherence tomography (OCT) has revolutionized the field of ophthalmology and vision science with its non-contact, non-invasive and high-resolution nature and become one of the prominent diagnostic tools in offering cross-sectional tissue imaging [14][15][16]. Applications of OCT for in vivo human eye imaging have been widely demonstrated in the literature [17][18][19][20][21][22][23][24][25][26][27][28]. With ever-improving techniques in broadband light sources, ultra-high resolution OCT (UHR-OCT) has achieved m...

show abstract

Section: Discussionmentioning

confidence: 99%

Multi-modal Anterior Eye Imager Combining Ultra-High Resolution OCT and Microvascular Imaging for Structural and Functional Evaluation of the Human Eye

et al. 2020

View full text Add to dashboard Cite

show abstract

“…Age-related macular degeneration (AMD) is one of the leading causes of blindness in the world [26]. Detectable AMD-related changes in OCTs are, among others, drusen, intra-and subretinal fluid, pigment epithelial detachment (PED) and photoreceptor loss [8]. Besides neovascular AMD, which is defined by the occurrence of fluid, geographic atrophy (GA) is the second form of late AMD, characterized by the death of retinal pigment epithelium (RPE) cells, photoreceptors and/or choriocapillaris.…”

Section: A Retinal Oct Imagingmentioning

confidence: 99%

“…Alternatively, supervised deep learning avoids biases due to manual design of features by learning them from data. These techniques have been extensively used to identify pre-defined pathological markers such as disease lesions [4], [7], [8], [34]. Their main drawback is that they require a training set with manual annotations of the region of interest.…”

Section: B Related Workmentioning

confidence: 99%

Exploiting Epistemic Uncertainty of Anatomy Segmentation for Anomaly Detection in Retinal OCT

Seeböck

Orlando

Schlegl

et al. 2020

IEEE Trans. Med. Imaging

Self Cite

134

View full text Add to dashboard Cite

Diagnosis and treatment guidance are aided by detecting relevant biomarkers in medical images. Although supervised deep learning can perform accurate segmentation of pathological areas, it is limited by requiring a-priori definitions of these regions, large-scale annotations, and a representative patient cohort in the training set. In contrast, anomaly detection is not limited to specific definitions of pathologies and allows for training on healthy samples without annotation. Anomalous regions can then serve as candidates for biomarker discovery. Knowledge about normal anatomical structure brings implicit information for detecting anomalies. We propose to take advantage of this property using bayesian deep learning, based on the assumption that epistemic uncertainties will correlate with anatomical deviations from a normal training set. A Bayesian U-Net is trained on a well-defined healthy environment using weak labels of healthy anatomy produced by existing methods. At test time, we capture epistemic uncertainty estimates of our model using Monte Carlo dropout. A novel post-processing technique is then applied to exploit these estimates and transfer their layered appearance to smooth blob-shaped segmentations of the anomalies. We experimentally validated this approach in retinal optical coherence tomography (OCT) images, using weak labels of retinal layers. Our method achieved a Dice index of 0.789 in an independent anomaly test set of age-related macular degeneration (AMD) cases. The resulting segmentations allowed very high accuracy for separating healthy and diseased cases with late wet AMD, dry geographic atrophy (GA), diabetic macular edema (DME) and retinal vein occlusion (RVO). Finally, we qualitatively observed that our approach can also detect other deviations in normal scans such as cut edge artifacts.

show abstract

“…Our experiments on a separate test set show that the model was robust enough to deal with different stages of the diseases that were not included on the training set. Hence, by incorporating uncertainty estimations and a modified U-Net architecture, we both improve the generalizability and the interpretability of our model during test time, favoring its application in clinical scenarios [20]. Further research will be performed to improve the results in areas of high uncertainty, and to correlate the uncertainty outcomes…”

Section: Discussionmentioning

confidence: 99%

U2-Net: A Bayesian U-Net Model With Epistemic Uncertainty Feedback For Photoreceptor Layer Segmentation In Pathological OCT Scans

Orlando

Seeböck

Bogunović

et al. 2019

2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019)

Self Cite

View full text Add to dashboard Cite

In this paper, we introduce a Bayesian deep learning based model for segmenting the photoreceptor layer in pathological OCT scans. Our architecture provides accurate segmentations of the photoreceptor layer and produces pixel-wise epistemic uncertainty maps that highlight potential areas of pathologies or segmentation errors. We empirically evaluated this approach in two sets of pathological OCT scans of patients with age-related macular degeneration, retinal vein oclussion and diabetic macular edema, improving the performance of the baseline U-Net both in terms of the Dice index and the area under the precision/recall curve. We also observed that the uncertainty estimates were inversely correlated with the model performance, underlying its utility for highlighting areas where manual inspection/correction might be needed.

show abstract

Artificial intelligence in retina

Cited by 585 publications

References 155 publications

Multi-modal Anterior Eye Imager Combining Ultra-High Resolution OCT and Microvascular Imaging for Structural and Functional Evaluation of the Human Eye

Multi-modal Anterior Eye Imager Combining Ultra-High Resolution OCT and Microvascular Imaging for Structural and Functional Evaluation of the Human Eye

Exploiting Epistemic Uncertainty of Anatomy Segmentation for Anomaly Detection in Retinal OCT

U2-Net: A Bayesian U-Net Model With Epistemic Uncertainty Feedback For Photoreceptor Layer Segmentation In Pathological OCT Scans

Contact Info

Product

Resources

About