High interpretable machine learning classifier for early glaucoma diagnosis

Escamez, Carlos Salvador Fernandez; Perucho-Martínez, S; Fernández, Nicolás Toledano

doi:10.18240/ijo.2021.03.10

Cited by 17 publications

(10 citation statements)

References 33 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Prior studies have also shown strong discriminative ability between healthy and glaucomatous eyes with these methods. 22 , 41 …”

Section: Discussionmentioning

confidence: 99%

Normative Percentiles of Retinal Nerve Fiber Layer Thickness and Glaucomatous Visual Field Loss

Singh,

Rauscher,

et al. 2023

Trans. Vis. Sci. Tech.

View full text Add to dashboard Cite

Purpose Circumpapillary retinal nerve fiber layer thickness (RNFLT) measurement aids in the clinical diagnosis of glaucoma. Spectral domain optical coherence tomography (SD-OCT) machines measure RNFLT and provide normative color-coded plots. In this retrospective study, we investigate whether normative percentiles of RNFLT (pRNFLT) from Spectralis SD-OCT improve prediction of glaucomatous visual field loss over raw RNFLT. Methods A longitudinal database containing OCT scans and visual fields from Massachusetts Eye & Ear glaucoma clinic patients was generated. Reliable OCT-visual field pairs were selected. Spectralis OCT normative distributions were extracted from machine printouts. Supervised machine learning models compared predictive performance between pRNFLT and raw RNFLT inputs. Regional structure-function associations were assessed with univariate regression to predict mean deviation (MD). Multivariable classification predicted MD, pattern standard deviation, MD change per year, and glaucoma hemifield test. Results There were 3016 OCT-visual field pairs that met the reliability criteria. Spectralis norms were found to be independent of age, sex, and ocular magnification. Regional analysis showed significant decrease in R 2 from pRNFLT models compared to raw RNFLT models in inferotemporal sectors, across multiple regressors. In multivariable classification, there were no significant improvements in area under the curve of receiver operating characteristic curve (ROC-AUC) score with pRNFLT models compared to raw RNFLT models. Conclusions Our results challenge the assumption that normative percentiles from OCT machines improve prediction of glaucomatous visual field loss. Raw RNFLT alone shows strong prediction, with no models presenting improvement by the manufacturer norms. This may result from insufficient patient stratification in tested norms. Translational Relevance Understanding correlation of normative databases to visual function may improve clinical interpretation of OCT data.

show abstract

“…Prior studies have also shown strong discriminative ability between healthy and glaucomatous eyes with these methods. 22 , 41 …”

Section: Discussionmentioning

confidence: 99%

Normative Percentiles of Retinal Nerve Fiber Layer Thickness and Glaucomatous Visual Field Loss

Singh,

Rauscher,

et al. 2023

Trans. Vis. Sci. Tech.

View full text Add to dashboard Cite

show abstract

“…In this review of related work, we will mainly focus on interpretability techniques applied to CNNs for glaucoma diagnosis, with special emphasis on the use of surrogate models. The use of tabular clinical data to fit interpretable machine learning models has been widely explored in glaucoma diagnosis, either alone [20,21] or in combination with fundus images [22][23][24]. Another common approach is the extraction of relevant features from images.…”

Section: Related Workmentioning

confidence: 99%

Interpretable surrogate models to approximate the predictions of convolutional neural networks in glaucoma diagnosis

Sigut,

Fumero,

Arnay

et al. 2023

Mach. Learn.: Sci. Technol.

View full text Add to dashboard Cite

Background and objectiveDeep learning systems, especially in critical fields like medicine, suffer from a significant drawback - their black box nature, which lacks mechanisms for explaining or interpreting their decisions. In this regard, our research aims to evaluate the use of surrogate models for interpreting convolutional neural network decisions in glaucoma diagnosis. Our approach is novel in that we approximate the original model with an interpretable one and also change the input features, replacing pixels with tabular geometric features of the optic disc, cup, and neuroretinal rim.MethodWe trained convolutional neural networks with two types of images: original images of the optic nerve head and simplified images showing only the disc and cup contours on a uniform background. Decision trees were used as surrogate models due to their simplicity and visualization properties, while saliency maps were calculated for some images for comparison.ResultsThe experiments carried out with 1271 images of healthy subjects and 721 images of glaucomatous eyes demonstrate that decision trees can closely approximate the predictions of neural networks trained on simplified contour images, with R-squared values near 0.9 for VGG19, Resnet50, InceptionV3 and Xception architectures. Saliency maps proved difficult to interpret and showed inconsistent results across architectures, in contrast to the decision trees. Additionally, some decision trees trained as surrogate models outperformed a decision tree trained on the actual outcomes without surrogation.ConclusionsDecision trees may be a more interpretable alternative to saliency methods. Moreover, the fact that we matched the performance of a decision tree without surrogation to that obtained by decision trees using knowledge distillation from neural networks is a great advantage since decision trees are inherently interpretable. Therefore, based on our findings, we think this approach would be the most recommendable choice for specialists as a diagnostic tool.

show abstract

“…For example, Wroblewski et al ( 2009 ) used support vector machines (SVMs) to provide a valid clinical diagnosis of glaucoma based solely on visual field data. Escamez et al ( 2021 ) developed a classifier for predicting glaucoma eyes based on peripapillary retinal nerve fiber layer (RNFL) thicknesses measured with OCT. The other is a multimodal fusion image, which is a combination of two or more types of data.…”

Section: Introductionsmentioning

confidence: 99%

A Multimodal Classification Architecture for the Severity Diagnosis of Glaucoma Based on Deep Learning

Zhang

Qian

et al. 2022

Front. Neurosci.

View full text Add to dashboard Cite

Glaucoma is an optic neuropathy that leads to characteristic visual field defects. However, there is no cure for glaucoma, so the diagnosis of its severity is essential for its prevention. In this paper, we propose a multimodal classification architecture based on deep learning for the severity diagnosis of glaucoma. In this architecture, a gray scale image of the visual field is first reconstructed with a higher resolution in the preprocessing stage, and more subtle feature information is provided for glaucoma diagnosis. We then use multimodal fusion technology to integrate fundus images and gray scale images of the visual field as the input of this architecture. Finally, the inherent limitation of convolutional neural networks (CNNs) is addressed by replacing the original classifier with the proposed classifier. Our architecture is trained and tested on the datasets provided by the First Affiliated Hospital of Kunming Medical University, and the results show that the proposed architecture achieves superior performance for glaucoma diagnosis.

show abstract

High interpretable machine learning classifier for early glaucoma diagnosis

Cited by 17 publications

References 33 publications

Normative Percentiles of Retinal Nerve Fiber Layer Thickness and Glaucomatous Visual Field Loss

Normative Percentiles of Retinal Nerve Fiber Layer Thickness and Glaucomatous Visual Field Loss

Interpretable surrogate models to approximate the predictions of convolutional neural networks in glaucoma diagnosis

A Multimodal Classification Architecture for the Severity Diagnosis of Glaucoma Based on Deep Learning

Contact Info

Product

Resources

About