Assessing the external validity of machine learning-based detection of glaucoma

Li, Chi; Chua, Jacqueline; Schwarzhans, Florian; Husain, Rahat; Girard, M; Majithia, Shivani; Tham, Yih‐Chung; Cheng, Ching‐Yu; Aung, Tin; Fischer, Georg; Vass, Clemens; Bujor, Inna; Kwoh, Chee Keong; Popa-Cherecheanu, Alina; Schmetterer, Leopold; Wong, Damon Wing Kee

doi:10.1038/s41598-023-27783-1

Cited by 15 publications

(7 citation statements)

References 40 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The number of subjects recruited in our study closely matches those from one of the most recent glaucoma and diabetic retinopathy clinical studies 39 . Indeed, there are some studies performed with PS-OCT that include more subjects, but these studies only performed experiments on patient eyes 40,41 .…”

Section: Discussionmentioning

confidence: 99%

Trans-ocular hypertension assessment

Afsharan

Anilkumar

Silva

et al. 2023

Preprint

View full text Add to dashboard Cite

Blood vessel walls are made of organized fibrous tissue with intrinsic birefringence. Even in its very early stages, hypertension can change the structure of a blood vessel wall. We showed that this structural change can be quantitatively measured non-invasively in the human retina using polarization-sensitive optical coherence tomography (PS-OCT). Organizational loss in vessel wall tissue was quantified in hypertensive patients and compared to data obtained from age-matched healthy subjects. The wall tissue in patients with hypertension was shown to be thicker, and exhibited lower birefringence, presumably due to a loss of tissue organization. The blood vessel wall birefringence index (BBI) combines vessel wall birefringence and thickness into one number and is introduced here to readily distinguish between blood vessel walls of hypertensive and healthy subjects. Based on receiver operating characteristic (ROC) curves, BBI showed 99% sensitivity and 100% specificity when discriminating normotensive (N = 11) and hypertensive (N = 11) subjects. Accurately determining the thickness of the blood vessel wall is not possible without polarization-sensitivity. Moreover, just blood vessel wall thickness or wall birefringence were not sufficient to achieve this high classification performance. Retinal vessel wall measurements with PS-OCT cannot be affected by a patient being anxious, by hormones or other blood values, since the measurement is intrinsic to the optical and mechanical properties of the vessel wall. Furthermore, this relatively low-cost system combines a very short imaging and analysis time with high sensitivity and specificity, making it highly suitable for low-cost screening.

show abstract

Section: Discussionmentioning

confidence: 99%

Trans-ocular hypertension assessment

Afsharan

Anilkumar

Silva

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…For the machine learning approaches, Li et al (2023) proposed a model for glaucoma detection using ML models with a two-class-labeled private dataset, and achieved accuracies of 79 and 84% for original and compensated datasets, respectively. Khan et al (2022) developed a glaucoma detection model using wavelet transform and ML model.…”

Section: Discussionmentioning

confidence: 99%

Multi-stage glaucoma classification using pre-trained convolutional neural networks and voting-based classifier fusion

Velpula,

Sharma

2023

Front. Physiol.

View full text Add to dashboard Cite

Aim: To design an automated glaucoma detection system for early detection of glaucoma using fundus images.Background: Glaucoma is a serious eye problem that can cause vision loss and even permanent blindness. Early detection and prevention are crucial for effective treatment. Traditional diagnostic approaches are time consuming, manual, and often inaccurate, thus making automated glaucoma diagnosis necessary.Objective: To propose an automated glaucoma stage classification model using pre-trained deep convolutional neural network (CNN) models and classifier fusion.Methods: The proposed model utilized five pre-trained CNN models: ResNet50, AlexNet, VGG19, DenseNet-201, and Inception-ResNet-v2. The model was tested using four public datasets: ACRIMA, RIM-ONE, Harvard Dataverse (HVD), and Drishti. Classifier fusion was created to merge the decisions of all CNN models using the maximum voting-based approach.Results: The proposed model achieved an area under the curve of 1 and an accuracy of 99.57% for the ACRIMA dataset. The HVD dataset had an area under the curve of 0.97 and an accuracy of 85.43%. The accuracy rates for Drishti and RIM-ONE were 90.55 and 94.95%, respectively. The experimental results showed that the proposed model performed better than the state-of-the-art methods in classifying glaucoma in its early stages. Understanding the model output includes both attribution-based methods such as activations and gradient class activation map and perturbation-based methods such as locally interpretable model-agnostic explanations and occlusion sensitivity, which generate heatmaps of various sections of an image for model prediction.Conclusion: The proposed automated glaucoma stage classification model using pre-trained CNN models and classifier fusion is an effective method for the early detection of glaucoma. The results indicate high accuracy rates and superior performance compared to the existing methods.

show abstract

“…The number of subjects recruited in our study closely matches those from one of the most recent glaucoma and diabetic retinopathy clinical studies [ 62 ]. Indeed, there are some studies performed with PS-OCT that include more subjects, but these studies only performed experiments on patient eyes [ 63 , 64 ]. However, a next study should include more patients with stages ranging from virgin diabetics or diabetic suspects to end-stage diabetics to understand the distribution of BBI at various stages of the disease.…”

Section: Discussionmentioning

confidence: 99%

Non-Invasive Retinal Blood Vessel Wall Measurements with Polarization-Sensitive Optical Coherence Tomography for Diabetes Assessment: A Quantitative Study

et al. 2023

View full text Add to dashboard Cite

Diabetes affects the structure of the blood vessel walls. Since the blood vessel walls are made of birefringent organized tissue, any change or damage to this organization can be evaluated using polarization-sensitive optical coherence tomography (PS-OCT). In this paper, we used PS-OCT along with the blood vessel wall birefringence index (BBI = thickness/birefringence2) to non-invasively assess the structural integrity of the human retinal blood vessel walls in patients with diabetes and compared the results to those of healthy subjects. PS-OCT measurements revealed that blood vessel walls of diabetic patients exhibit a much higher birefringence while having the same wall thickness and therefore lower BBI values. Applying BBI to diagnose diabetes demonstrated high accuracy (93%), sensitivity (93%) and specificity (93%). PS-OCT measurements can quantify small changes in the polarization properties of retinal vessel walls associated with diabetes, which provides researchers with a new imaging tool to determine the effects of exercise, medication, and alternative diets on the development of diabetes.

show abstract

Assessing the external validity of machine learning-based detection of glaucoma

Cited by 15 publications

References 40 publications

Trans-ocular hypertension assessment

Trans-ocular hypertension assessment

Multi-stage glaucoma classification using pre-trained convolutional neural networks and voting-based classifier fusion

Non-Invasive Retinal Blood Vessel Wall Measurements with Polarization-Sensitive Optical Coherence Tomography for Diabetes Assessment: A Quantitative Study

Contact Info

Product

Resources

About