Evaluating the Impact of Retinal Vessel Segmentation Metrics on Retest Reliability in a Clinical Setting: A Comparative Analysis Using AutoMorph DOI Creative Commons

Samuel David Giesser,

Ferhat Turgut, Amr Saad

et al.

Investigative Ophthalmology & Visual Science, Journal Year: 2024, Volume and Issue: 65(13), P. 24 - 24

Published: Nov. 14, 2024

Purpose: Current research on artificial intelligence–based fundus photography biomarkers has demonstrated inconsistent results. Consequently, we aimed to evaluate and predict the test–retest reliability of retinal parameters extracted from photography. Methods: Two groups patients were recruited for study: an intervisit group (n = 28) assess retest over a period 1 14 days intravisit 44) within single session. Using AutoMorph, generated test vessel segmentation maps; measured map agreement via accuracy, sensitivity, F1 score Jaccard index; calculated 76 metrics each image. The metric was analyzed in terms Spearman correlation coefficient, intraclass coefficient (ICC), relative percentage change. A linear model with input variables contrast-to-noise-ratio fractal dimension, chosen by P-value–based backward selection process, developed median difference per image based image-quality metrics. This trained dataset validated using dataset. Results: In group, varied between coefficients 0.34 0.99, ICC values 0.31 mean absolute differences 0.96% 223.67%. Similarly, ranged 0.55 0.96, 0.40 0.97, 0.49% 371.23%. Segmentation accuracy never dropped below 97%; scores 0.85 0.82 best achieved disc-width regarding both datasets. worst retests tortuosity density artery density, respectively. exhibited better than (P < 0.001). Our model, two independent contrast-to-noise ratio dimension predicted its validation dataset, R2 0.53 Conclusions: findings highlight considerable volatility some biomarkers. Improving could allow disease progression modeling smaller datasets or individualized treatment approach. Image quality is moderately predictive reliability, further work warranted understand reasons behind our observations thus ensure consistent

Language: Английский

A Retinal Oct-Angiography and Cardiovascular STAtus (RASTA) Dataset of Swept-Source Microvascular Imaging for Cardiovascular Risk Assessment DOI Creative Commons

Clément Germanèse,

Fabrice Mériaudeau, Pétra Eid

et al.

Data, Journal Year: 2023, Volume and Issue: 8(10), P. 147 - 147

Published: Sept. 28, 2023

In the context of exponential demographic growth, imbalance between human resources and public health problems impels us to envision other solutions difficulties faced in diagnosis, prevention, large-scale management most common diseases. Cardiovascular diseases represent leading cause morbidity mortality worldwide. A screening program would make it possible promptly identify patients with high cardiovascular risk order manage them adequately. Optical coherence tomography angiography (OCT-A), as a window into state system, is rapid, reliable, reproducible imaging examination that enables prompt identification at-risk through use automated classification models. One challenge limits development computer-aided diagnostic programs small number open-source OCT-A acquisitions available. To facilitate such models, we have assembled set images retinal microvascular system from 499 patients. It consists 814 angiocubes well 2005 en face images. Angiocubes were captured swept-source device varying overall risk. best our knowledge, dataset, Retinal oct-Angiography STAtus (RASTA), only publicly available dataset comprising variety healthy This will enable generalizable models for

Language: Английский

Citations

5

Artificial intelligence-based prediction of neurocardiovascular risk score from retinal swept-source microvascular imaging: the RASTA dataset DOI Creative Commons

Clément Germanèse,

Atif Anwer, Pétra Eid

et al.

Research Square (Research Square), Journal Year: 2024, Volume and Issue: unknown

Published: May 2, 2024

Abstract The recent rise of artificial intelligence represents a revolutionary way improving current medical practices, including cardiovascular (CV) assessment scores. Retinal vascular alterations may reflect systemic processes such as the presence CV risk factors. value swept-source retinal optical coherence tomography–angiography (SS OCT-A) imaging is significantly enhanced by image analysis tools that provide rapid and accurate quantification features. We report on interest using machine-learning (ML) deep-learning (DL) models for from SS OCT-A microvasculature imaging. assessed accuracy ML DL algorithms in predicting CHA2DS2-VASc neurocardiovascular score based images patients open-source RASTA dataset. were trained data 491 patients. tested here achieved good performance with area under curve (AUC) values ranging 0.71 to 0.96. According classification into two or three groups, EfficientNetV2-B3 tool predicted correctly 39% 68% cases, respectively, mean absolute error (MAE) approximately 0.697. Our enable confident prediction imaging, which could be useful contributing profiles future.

Language: Английский

Citations

1

A Multi-Stage Approach for Cardiovascular Risk Assessment from Retinal Images Using an Amalgamation of Deep Learning and Computer Vision Techniques DOI Creative Commons
Deepthi K Prasad, Madhura Prakash M,

Meghna S Kulkarni

et al.

Diagnostics, Journal Year: 2024, Volume and Issue: 14(9), P. 928 - 928

Published: April 29, 2024

Cardiovascular diseases (CVDs) are a leading cause of mortality worldwide. Early detection and effective risk assessment crucial for implementing preventive measures improving patient outcomes CVDs. This work presents novel approach to CVD using fundus images, leveraging the inherent connection between retinal microvascular changes systemic vascular health. study aims develop predictive model early CVDs by evaluating parameters. methodology integrates both handcrafted features derived through mathematical computation patterns extracted artificial intelligence (AI) models. By combining these approaches, we seek enhance accuracy reliability prediction in individuals. The state-of-the-art computer vision algorithms AI techniques multi-stage architecture extract relevant from images. These encompass range parameters, including vessel caliber, tortuosity, branching patterns. Additionally, deep learning (DL)-based binary classification is incorporated accuracy. A dataset comprising images comprehensive metadata clinical trials conducted utilized training validation. proposed demonstrates promising results factors. interpretability enhanced visualization that highlight regions interest within contributing predictions. Furthermore, validation performance analysis shows potential provide accurate system not only aids stratification but also serves as valuable tool identifying abnormalities may precede overt cardiovascular events. has achieved an 85% findings this underscore feasibility efficacy assessment. As non-invasive cost-effective modality, image scalable solution population-wide screening programs. research contributes evolving landscape precision medicine providing innovative proactive health management. Future will focus on refining solution’s robustness, exploring additional factors, validating its diverse settings.

Language: Английский

Citations

1

Development of oculomics artificial intelligence for cardiovascular risk factors: A case study in fundus oculomics for HbA1c assessment and clinically relevant considerations for clinicians DOI Creative Commons
Joshua Ong, Kuk Jin Jang,

Seung Ju Baek

et al.

Asia-Pacific Journal of Ophthalmology, Journal Year: 2024, Volume and Issue: 13(4), P. 100095 - 100095

Published: July 1, 2024

Artificial Intelligence (AI) is transforming healthcare, notably in ophthalmology, where its ability to interpret images and data can significantly enhance disease diagnosis patient care. Recent developments oculomics, the integration of ophthalmic features develop biomarkers for systemic diseases, have demonstrated potential providing rapid, non-invasive methods screening leading early detection improve healthcare quality, particularly underserved areas. However, widespread adoption such AI-based technologies faces challenges primarily related trustworthiness system. We demonstrate considerations needed trustworthy AI oculomics through a pilot study HbA1c assessment using an approach. then discuss various challenges, considerations, solutions that been developed powerful past subsequently apply these study. Building upon observations we highlight opportunities advancing oculomics. Ultimately, presents as emerging technology ophthalmology understanding how optimize transparency prior clinical utmost importance.

Language: Английский

Citations

1

Evaluating the Impact of Retinal Vessel Segmentation Metrics on Retest Reliability in a Clinical Setting: A Comparative Analysis Using AutoMorph DOI Creative Commons

Samuel David Giesser,

Ferhat Turgut, Amr Saad

et al.

Investigative Ophthalmology & Visual Science, Journal Year: 2024, Volume and Issue: 65(13), P. 24 - 24

Published: Nov. 14, 2024

Purpose: Current research on artificial intelligence–based fundus photography biomarkers has demonstrated inconsistent results. Consequently, we aimed to evaluate and predict the test–retest reliability of retinal parameters extracted from photography. Methods: Two groups patients were recruited for study: an intervisit group (n = 28) assess retest over a period 1 14 days intravisit 44) within single session. Using AutoMorph, generated test vessel segmentation maps; measured map agreement via accuracy, sensitivity, F1 score Jaccard index; calculated 76 metrics each image. The metric was analyzed in terms Spearman correlation coefficient, intraclass coefficient (ICC), relative percentage change. A linear model with input variables contrast-to-noise-ratio fractal dimension, chosen by P-value–based backward selection process, developed median difference per image based image-quality metrics. This trained dataset validated using dataset. Results: In group, varied between coefficients 0.34 0.99, ICC values 0.31 mean absolute differences 0.96% 223.67%. Similarly, ranged 0.55 0.96, 0.40 0.97, 0.49% 371.23%. Segmentation accuracy never dropped below 97%; scores 0.85 0.82 best achieved disc-width regarding both datasets. worst retests tortuosity density artery density, respectively. exhibited better than (P < 0.001). Our model, two independent contrast-to-noise ratio dimension predicted its validation dataset, R2 0.53 Conclusions: findings highlight considerable volatility some biomarkers. Improving could allow disease progression modeling smaller datasets or individualized treatment approach. Image quality is moderately predictive reliability, further work warranted understand reasons behind our observations thus ensure consistent

Language: Английский

Citations

1