Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Advanced Assessment of Stroke in Retinal Fundus Imaging with Deep Multi-view Learning

Aysen Degerli, Mika Hilvo, Juha Pajula, Petri Huhtinen, Pekka Jäkälä

TL;DR

This work introduces MVS-Net, a deep multi-view network that processes four retinal fundus images—macula- and optic nerve head-centered views from both eyes—to detect stroke and discriminate from TIA and healthy controls. By leveraging four input channels and backbone CNNs pretrained on ImageNet, the model achieves an AUC of up to $0.84$ for binary stroke detection, outperforming single-view baselines and illustrating the benefits of dual-view, cross-eye information. The Stroke-Data dataset, comprising 802 images from 220 subjects, enables the first inclusion of TIA in this context, highlighting challenges in identifying TIA and the limitations imposed by dataset size and class imbalance. The findings support potential deployment of MVS-Net in portable fundus cameras to aid rapid stroke triage, while underscoring the need for larger, more diverse datasets and external validation for clinical translation.

Abstract

Stroke is globally a major cause of mortality and morbidity, and hence accurate and rapid diagnosis of stroke is valuable. Retinal fundus imaging reveals the known markers of elevated stroke risk in the eyes, which are retinal venular widening, arteriolar narrowing, and increased tortuosity. In contrast to other imaging techniques used for stroke diagnosis, the acquisition of fundus images is easy, non-invasive, fast, and inexpensive. Therefore, in this study, we propose a multi-view stroke network (MVS-Net) to detect stroke and transient ischemic attack (TIA) using retinal fundus images. Contrary to existing studies, our study proposes for the first time a solution to discriminate stroke and TIA with deep multi-view learning by proposing an end-to-end deep network, consisting of multi-view inputs of fundus images captured from both right and left eyes. Accordingly, the proposed MVS-Net defines representative features from fundus images of both eyes and determines the relation within their macula-centered and optic nerve head-centered views. Experiments performed on a dataset collected from stroke and TIA patients, in addition to healthy controls, show that the proposed framework achieves an AUC score of 0.84 for stroke and TIA detection.

Advanced Assessment of Stroke in Retinal Fundus Imaging with Deep Multi-view Learning

TL;DR

This work introduces MVS-Net, a deep multi-view network that processes four retinal fundus images—macula- and optic nerve head-centered views from both eyes—to detect stroke and discriminate from TIA and healthy controls. By leveraging four input channels and backbone CNNs pretrained on ImageNet, the model achieves an AUC of up to for binary stroke detection, outperforming single-view baselines and illustrating the benefits of dual-view, cross-eye information. The Stroke-Data dataset, comprising 802 images from 220 subjects, enables the first inclusion of TIA in this context, highlighting challenges in identifying TIA and the limitations imposed by dataset size and class imbalance. The findings support potential deployment of MVS-Net in portable fundus cameras to aid rapid stroke triage, while underscoring the need for larger, more diverse datasets and external validation for clinical translation.

Abstract

Stroke is globally a major cause of mortality and morbidity, and hence accurate and rapid diagnosis of stroke is valuable. Retinal fundus imaging reveals the known markers of elevated stroke risk in the eyes, which are retinal venular widening, arteriolar narrowing, and increased tortuosity. In contrast to other imaging techniques used for stroke diagnosis, the acquisition of fundus images is easy, non-invasive, fast, and inexpensive. Therefore, in this study, we propose a multi-view stroke network (MVS-Net) to detect stroke and transient ischemic attack (TIA) using retinal fundus images. Contrary to existing studies, our study proposes for the first time a solution to discriminate stroke and TIA with deep multi-view learning by proposing an end-to-end deep network, consisting of multi-view inputs of fundus images captured from both right and left eyes. Accordingly, the proposed MVS-Net defines representative features from fundus images of both eyes and determines the relation within their macula-centered and optic nerve head-centered views. Experiments performed on a dataset collected from stroke and TIA patients, in addition to healthy controls, show that the proposed framework achieves an AUC score of 0.84 for stroke and TIA detection.

Paper Structure

This paper contains 10 sections, 4 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Materials and Methods
  3. Dataset
  4. Multi-view Stroke Network
  5. Model Structure
  6. Model Training
  7. Evaluation Metrics
  8. Results
  9. Discussion
  10. Conclusion

Figures (4)

  • Figure 1: The proposed multi-view stroke network (MVS-Net) is illustrated, where dual-view retinal fundus images are given to MSV-Net as inputs to predict stroke, TIA, and healthy controls.
  • Figure 2: Confusion matrices of the MVS-Net, where the backbone is (a) ResNet50, (b) Xception, and (c) Inception-v3 models.
  • Figure 3: The model training curves of the proposed MVS-Net are plotted with respect to train accuracy and loss, when the network is trained for discriminating stroke, TIA, and healthy controls. The backbone of the proposed model is indicated in the subfigure titles.
  • Figure 4: Radar charts of MVS-Net with ResNet50 (a) and VGG19 (b) backbone models for comparison to existing methods.