Automated algorithms combining structure and function outperform general ophthalmologists in diagnosing glaucoma

Shigueoka, Leonardo; Vasconcellos, José Paulo Cabral de; Schimiti, Rui Barroso; Reis, Arlete Barbosa dos; Oliveira, Gabriel Ozeas de; Gomi, Edson Satoshi; Vianna, Jayme R; Lisboa, Renato; Medeiros, Felipe A.; Costa, Vital Paulino

doi:10.1371/journal.pone.0207784

Cited by 33 publications

(21 citation statements)

References 44 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…In addition, the diagnostic accuracy of general ophthalmologists (0.8195) was lower than MLCs. Interestingly, studies comparing diagnosing ability between AI and ophthalmologists often shows better performance of AI, which are similar to our results [37,47]. Therefore, we can predict that in the future application of machine diagnosis, it can help general ophthalmologists and primary hospitals to make more accurate prediction and avoid missed diagnosis or misdiagnosis to some extent.…”

Section: Discussionsupporting

confidence: 88%

“…Previous attempts have been made to combine structure and function to detect POAG. In the studies of Raza et al[36] and Leonardo et al [37], researchers combined OCT and SAP results, and the results obtained were better than that obtained by analyzing OCT results alone, indicating that structure and function combining models show better diagnostic results. Kim et al…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

Multimodal Primary Open Angle Glaucoma Early Diagnosing Program Based on Clinical Process

et al. 2021

Preprint

View full text Add to dashboard Cite

Background: Glaucoma is one of the leading causes of blinding disease. Early detection can improve patients’ quality of vision. Effectively identifying primary open angle glaucoma (POAG) using structural and functional examination is critical. Computer aided diagnosis of glaucoma requires multimodal data to find an accurate model for early glaucoma diagnosis. Methods: This study collected 87 early POAG eyes, 85 suspected POAG eyes, and 129 healthy eyes from the ophthalmology department at Second Affiliated Hospital of Harbin Medical University. Retinal nerve fiber layer thickness (RNFLt), intraocular pressure (IOP) value, visual field examination parameters and age were obtained. A powerful deep learning network segmented the FP and extracted optic nerve head (ONH) features. Machine learning classifiers (MLCs) were adopted to get the final classification results and compared with the diagnosis results of glaucoma specialists and general non-glaucoma ophthalmologists. Result: The program diagnosing early POAG, suspected POAG, and healthy eyes made overall Area Under the Curve of 0.97. Dice of optic disc and optic cup segmentation is 0.9631, 0.8435 respectively. Accuracy of the program (0.9004) is higher than general ophthalmologists (0.8195). Specificity of the program (0.9635) is higher than glaucoma specialists (0.9366).Conclusions: The program delivers superior results in diagnosing early POAG. This study’s hybrid deep learning-machine learning framework can assist with clinical decision for early POAG effectively.

show abstract

Section: Discussionsupporting

confidence: 88%

Section: Discussionmentioning

confidence: 99%

Multimodal Primary Open Angle Glaucoma Early Diagnosing Program Based on Clinical Process

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…Further, clinicians are encouraged to confirm a suspected structural lesion in the functional domain to increase confidence in diagnosis. [16][17][18][19] Enface images offer the opportunity to relate structure and function without use of structure-function maps, thought to represent an additional source of noise in this relationship. 20 Indeed, newer perimetric strategies that aim to incorporate structural information for greater efficiency, [21][22][23] or that assess specific regions of interest in greater detail, [24][25][26] may be facilitated by enface imaging.…”

Section: Introductionmentioning

confidence: 99%

“…It has been suggested that disruption of RNFB reflectivity may be measurable earlier than RNFL thinning in glaucoma, 8,15 making this approach encouraging for earlier identification of defects. Further, clinicians are encouraged to confirm a suspected structural lesion in the functional domain to increase confidence in diagnosis 16–19 . Enface images offer the opportunity to relate structure and function without use of structure‐function maps, thought to represent an additional source of noise in this relationship 20 .…”

Section: Introductionmentioning

confidence: 99%

Depth‐resolved variations in visibility of retinal nerve fibre bundles across the retina in enface OCT images of healthy eyes

Cheloni

Denniss

2020

Ophthalmic Physiologic Optic

View full text Add to dashboard Cite

Purpose Recent developments in optical coherence tomography (OCT) technology enable direct enface visualisation of retinal nerve fibre bundle (RNFB) loss in glaucoma. However, the optimum depth at which to visualise RNFBs across the retina is unknown. We aimed to evaluate the range of depths and optimum depth at which RNFBs can be visualised across the retina in healthy eyes. Methods The central ± 25° retina of 10 healthy eyes from 10 people aged 57–75 years (median 68.5 years) were imaged with spectral domain OCT. Slab images of maximum axial resolution (4 μm) containing depth‐resolved attenuation coefficients were extracted from 0 to 193.5 μm below the inner limiting membrane (ILM). Bundle visibility within 10 regions of a superimposed grid was assessed subjectively by trained optometrists (n = 8), according to written instructions. Anterior and posterior limits of RNFB visibility and depth of best visibility were identified for each grid sector. Effects of retinal location and individual eye on RNFB visibility were explored using linear mixed modelling with likelihood ratio tests. Intraclass correlation coefficient (ICC) was used to measure overall agreement and repeatability of grading. Spearman’s correlation was used to measure correlation between depth range of visible RNFBs and retinal nerve fibre layer thickness (RNFLT). Results Retinal location and individual eye affected anterior limit of visibility (χ2(9) = 58.6 and 60.5, both p < 0.0001), but none of the differences exceeded instrument resolution, making anterior limit consistent across the retina and different eyes. Greater differences were observed in the posterior limit of visibility across retinal areas (χ2(9) = 1671.1, p < 0.0001) and different eyes (χ2(9) = 88.7, p < 0.0001). Optimal depth for visualisation of RNFBs was around 20 µm below the ILM in most regions. It varied slightly with retinal location (χ2(9) = 58.8, p < 0.0001), but it was not affected by individual eye (χ2(9) = 10.7, p = 0.29). RNFB visibility showed good agreement between graders (ICC 0.89, 95%CI 0.87–0.91), and excellent repeatability (ICC 0.96–0.99). Depth range of visible RNFBs was highly correlated with RNFLT (ρ = 0.9, 95%CI: 0.86–0.95). Conclusions The range of depths with visible RNFBs varies markedly across the healthy retina, consistently with RNFLT. To extract all RNFB information consistently across the retina, slab properties should account for differences across retinal locations and between individual eyes.

show abstract

“…As an example, previous studies have used SVMs to improve detection of glaucoma damage from imaging data. [11][12][13] The SVMs used features such as global and sectoral parameters of retinal nerve fiber layer (RNFL) thickness, and measurements such as rim and cup area, cleverly combining them to reach a final glaucoma classification. However, although satisfactory performance has been reported for SVMs and other traditional algorithms in this scenario, there is no guarantee that the parameters used as initial features make the best use of the vast information produced by imaging.…”

Section: Artificial Intelligence Machine Learning and Deep Learningmentioning

confidence: 99%

A Review of Deep Learning for Screening, Diagnosis, and Detection of Glaucoma Progression

Thompson

Jammal

Medeiros

2020

Trans. Vis. Sci. Tech.

121

View full text Add to dashboard Cite

Because of recent advances in computing technology and the availability of large datasets, deep learning has risen to the forefront of artificial intelligence, with performances that often equal, or sometimes even exceed, those of human subjects on a variety of tasks, especially those related to image classification and pattern recognition. As one of the medical fields that is highly dependent on ancillary imaging tests, ophthalmology has been in a prime position to witness the application of deep learning algorithms that can help analyze the vast amount of data coming from those tests. In particular, glaucoma stands as one of the conditions where application of deep learning algorithms could potentially lead to better use of the vast amount of information coming from structural and functional tests evaluating the optic nerve and macula. The purpose of this article is to critically review recent applications of deep learning models in glaucoma, discussing their advantages but also focusing on the challenges inherent to the development of such models for screening, diagnosis and detection of progression. After a brief general overview of deep learning and how it compares to traditional machine learning classifiers, we discuss issues related to the training and validation of deep learning models and how they specifically apply to glaucoma. We then discuss specific scenarios where deep learning has been proposed for use in glaucoma, such as screening with fundus photography, and diagnosis and detection of glaucoma progression with optical coherence tomography and standard automated perimetry. Translational Relevance Deep learning algorithms have the potential to significantly improve diagnostic capabilities in glaucoma, but their application in clinical practice requires careful validation, with consideration of the target population, the reference standards used to build the models, and potential sources of bias.

show abstract

Automated algorithms combining structure and function outperform general ophthalmologists in diagnosing glaucoma

Cited by 33 publications

References 44 publications

Multimodal Primary Open Angle Glaucoma Early Diagnosing Program Based on Clinical Process

Multimodal Primary Open Angle Glaucoma Early Diagnosing Program Based on Clinical Process

Depth‐resolved variations in visibility of retinal nerve fibre bundles across the retina in enface OCT images of healthy eyes

A Review of Deep Learning for Screening, Diagnosis, and Detection of Glaucoma Progression

Contact Info

Product

Resources

About