Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Predicting risk of cardiovascular disease using retinal OCT imaging

Cynthia Maldonado-Garcia, Rodrigo Bonazzola, Enzo Ferrante, Thomas H Julian, Panagiotis I Sergouniotis, Nishant Ravikumara, Alejandro F Frangi

TL;DR

This study presents a non-invasive approach to predicting cardiovascular events within five years by coupling self-supervised learning on 3D retinal OCT data with a multimodal Random Forest classifier that also incorporates basic clinical metadata. The model learns latent eye-specific representations via independent variational autoencoders and fuses them with metadata, achieving an AUC of $0.75$ and accuracy of $0.70$, surpassing the QRISK3 baseline (AUC $0.60$, accuracy $0.55$). A key finding is the prominence of the choroidal layer and left-eye features in predicting future CVD events, with a novel optical-flow–based explainability method providing both local and global insights into retinal regions driving predictions. The results support OCT-based screening as a cost-effective, scalable supplement to existing cardiovascular risk tools, with potential integration into optometry and clinical settings for early risk stratification.

Abstract

Cardiovascular diseases (CVD) are the leading cause of death globally. Non-invasive, cost-effective imaging techniques play a crucial role in early detection and prevention of CVD. Optical coherence tomography (OCT) has gained recognition as a potential tool for early CVD risk prediction, though its use remains underexplored. In this study, we investigated the potential of OCT as an additional imaging technique to predict future CVD events. We analysed retinal OCT data from the UK Biobank. The dataset included 612 patients who suffered a myocardial infarction (MI) or stroke within five years of imaging and 2,234 controls without CVD (total: 2,846 participants). A self-supervised deep learning approach based on Variational Autoencoders (VAE) was used to extract low-dimensional latent representations from high-dimensional 3D OCT images, capturing distinct features of retinal layers. These latent features, along with clinical data, were used to train a Random Forest (RF) classifier to differentiate between patients at risk of future CVD events (MI or stroke) and healthy controls. Our model achieved an AUC of 0.75, sensitivity of 0.70, specificity of 0.70, and accuracy of 0.70, outperforming the QRISK3 score (the third version of the QRISK cardiovascular disease risk prediction algorithm; AUC = 0.60, sensitivity = 0.60, specificity = 0.55, accuracy = 0.55). The choroidal layer in OCT images was identified as a key predictor of future CVD events, revealed through a novel model explainability approach. This study demonstrates that retinal OCT imaging is a cost-effective, non-invasive alternative for predicting CVD risk, offering potential for widespread application in optometry practices and hospitals.

Predicting risk of cardiovascular disease using retinal OCT imaging

TL;DR

This study presents a non-invasive approach to predicting cardiovascular events within five years by coupling self-supervised learning on 3D retinal OCT data with a multimodal Random Forest classifier that also incorporates basic clinical metadata. The model learns latent eye-specific representations via independent variational autoencoders and fuses them with metadata, achieving an AUC of and accuracy of , surpassing the QRISK3 baseline (AUC , accuracy ). A key finding is the prominence of the choroidal layer and left-eye features in predicting future CVD events, with a novel optical-flow–based explainability method providing both local and global insights into retinal regions driving predictions. The results support OCT-based screening as a cost-effective, scalable supplement to existing cardiovascular risk tools, with potential integration into optometry and clinical settings for early risk stratification.

Abstract

Cardiovascular diseases (CVD) are the leading cause of death globally. Non-invasive, cost-effective imaging techniques play a crucial role in early detection and prevention of CVD. Optical coherence tomography (OCT) has gained recognition as a potential tool for early CVD risk prediction, though its use remains underexplored. In this study, we investigated the potential of OCT as an additional imaging technique to predict future CVD events. We analysed retinal OCT data from the UK Biobank. The dataset included 612 patients who suffered a myocardial infarction (MI) or stroke within five years of imaging and 2,234 controls without CVD (total: 2,846 participants). A self-supervised deep learning approach based on Variational Autoencoders (VAE) was used to extract low-dimensional latent representations from high-dimensional 3D OCT images, capturing distinct features of retinal layers. These latent features, along with clinical data, were used to train a Random Forest (RF) classifier to differentiate between patients at risk of future CVD events (MI or stroke) and healthy controls. Our model achieved an AUC of 0.75, sensitivity of 0.70, specificity of 0.70, and accuracy of 0.70, outperforming the QRISK3 score (the third version of the QRISK cardiovascular disease risk prediction algorithm; AUC = 0.60, sensitivity = 0.60, specificity = 0.55, accuracy = 0.55). The choroidal layer in OCT images was identified as a key predictor of future CVD events, revealed through a novel model explainability approach. This study demonstrates that retinal OCT imaging is a cost-effective, non-invasive alternative for predicting CVD risk, offering potential for widespread application in optometry practices and hospitals.
Paper Structure (14 sections, 4 equations, 10 figures, 8 tables)

This paper contains 14 sections, 4 equations, 10 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Database
  4. Framework of Self-Supervised Feature Selection Variational Autoencoder and Multimodal Random Forest Classification
  5. Self-supervised VAE
  6. Classification
  7. Experiment Details
  8. Model Explainability
  9. Results
  10. Classification Performance
  11. Model Explainability
  12. Discussion
  13. Conclusion
  14. Supplementary Figures and Tables

Figures (10)

  • Figure 1: STROBE flow chart describing participant inclusion and exclusion criteria applied to define the study cohort.
  • Figure 2: Distribution of age-sex cohort match. The left histogram illustrates the total of CVD+ labeled data used solely for the classification task. The middle histogram shows the total of CVD- for the classification task, while the right histogram illustrates CVD- used for the pretraining task.
  • Figure 3: The provided workflow diagram illustrates the comprehensive process of training the Variational Autoencoder (VAE) and subsequently using it to acquire the latent vectors (upper section). These latent vectors are then combined with metadata and serve as inputs to the Random Forest (RF) classifier (middle section). Finally, we perform an interpretability analysis by perturbing the most relevant features, reconstructing the corresponding image and computing the optical flow between the perturbed reconstructions (lower section). $z_{left}$ represents the latent vector obtained from the training of the VAE for the left eye. $Z_{right}$ corresponds to the latent vector acquired from training the VAE for the right eye.
  • Figure 4: Comparison of classifier performance in terms of True Positives (TP), True Negatives (TN), False Positives (FP), and False Negatives (FN). Values above the bars represent the corresponding counts for each classifier.
  • Figure 5: Calculation of feature importance magnitudes for the seven different classifiers investigated, where each classifier uses different combinations of data channels/modalities.
  • ...and 5 more figures