Advanced Assessment of Stroke in Retinal Fundus Imaging with Deep Multi-view Learning
Aysen Degerli, Mika Hilvo, Juha Pajula, Petri Huhtinen, Pekka Jäkälä · 2025
Read Advanced Assessment of Stroke in Retinal Fundus Imaging with Deep Multi-view Learning by Aysen Degerli, Mika Hilvo, Juha Pajula, Petri Huhtinen, Pekka Jäkälä (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-00079.txt.
Introduction
The back of the eye shows blood vessel changes that track stroke risk, and a photo of it is quick, cheap, and painless. Aysen Degerli and co-authors build a deep learning system in a 2025 paper that reads those photos for stroke and mini-stroke. Their network looks at both eyes at once, which earlier work had not done.
The system takes views centered on the macula and on the optic nerve head from each eye. It learns features from all four views and weighs how they relate. Tested on patients with stroke or mini-stroke plus healthy controls, it reached a solid discrimination score. The authors pitch eye photos as a fast front door for stroke checks where brain scans are scarce.
Catalog record
| CREATOR | Aysen Degerli, Mika Hilvo, Juha Pajula, Petri Huhtinen, Pekka Jäkälä |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 34 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-00079
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Advanced Assessment of Stroke in Retinal Fundus Imaging with Deep Multi-view Learning Aysen Degerli, Mika Hilvo, Juha Pajula, Petri Huhtinen, Pekka Jäkälä 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. I Introduction Stroke is a severe and prevalent manifestation of cerebrovascular disease that occurs due to an occluded or ruptured artery supplying blood to the brain [ 1 ] . World Health Organization (WHO) recognizes stroke as the second leading cause of death and disability [ 2 ] . According to the statistics reported by the World Stroke Organization (WSO), annually 12.2 million new strokes occur, and 6.5 million deaths are caused by stroke [ 3 ] . Although age is the most significant demographic risk factor, 16\% of strokes are encountered at the age of 15-49 , resulting in mortality and morbidity due to stroke [ 3 , 4 ] . Henceforth, the global cost for the diagnosis and treatment of stroke sharply increases as it is no longer confined to be primarily observed in the elderly population. Therefore, accurate and immediate diagnosis of stroke plays a significant role in the acceleration of its treatment and prevention of mortality and disability. Cerebral vasculature is formed by the vessels maintaining the blood supply and flow to the brain [ 5 ] . Cerebrovascular disease (CVD) encompasses disorders or diseases of the cerebral blood vessels, blood flow, and oxygenation [ 5 , 6 ] . Stroke is the most severe clinical manifestation among CVDs [ 7 ] . The main types of stroke are categorized as ischemic stroke, hemorrhagic stroke, and transient ischemic attack (TIA). 71\% of strokes are ischemic strokes that cause infarction in the brain, spinal cord, or retina [ 2 ] . Contrary to ischemic stroke, hemorrhagic stroke is a rare but more vital case that occurs when the artery supplying blood to the brain ruptures [ 8 ] . According to the stroke definition of WHO, ischemic stroke symptoms continue over \geq 24 hours or even lead to death [ 9 ] . In comparison to ischemic stroke, TIAs last less than 24 hours and resolve before leaving any permanent damage [ 10 ] . Hence, TIA is called a mini-stroke caused by the temporary interruption of blood flow to the brain [ 11 ] . Of note, TIA results in a 20-25\% risk of future ischemic stroke as the pathogenesis of TIA and ischemic stroke is identical [ 10 ] . However, discriminating TIA from non-cerebrovascular diseases is a challenging task, and 30\%-50\% of patients are misdiagnosed [ 12 ] . Therefore, the diagnosis of TIA is vital to prevent a possible future ischemic stroke. Pre-hospital stroke scale tests, electroencephalography (EEG), and imaging techniques are used in the initial diagnosis of a stroke. Accordingly, pre-hospital stroke scale tests such as the National Institutes of Health Stroke Scale (NIHSS), Face Arm Speech Time (F.A.S.T.), and Cincinnati Pre-hospital Stroke Scale (CPSS) are primary tests for detecting the presence of a stroke, which are used to assess the facial expressions, motor skills, and any speech disorder [ 13 ] . However, these tests face challenges due to the heterogeneous nature of stroke, which may result in unreliability, bias, and inability to manage time in diagnosing a stroke [ 14 , 15 ] . On the other hand, EEG enables monitoring of the electrical activity of the brain, which provides valuable information for the diagnosis of various neurological disorders, including stroke. However, EEG signals are interpreted more efficiently in stroke diagnosis when utilized together with other imaging modalities [ 16 ] . Hence, imaging techniques are crucial for the initial stroke identification, since brain infarction is typically monitored using Computed Tomography (CT) or Magnetic Resonance Imaging (MRI) [ 17 ] . Considering the time, limited sensitivity, and cost of acquiring conventional CT and MRI images and radiation exposure in CT imaging, alternative imaging modalities are of great interest in the biomedical field [ 18 , 19 ] . Retina can reflect the elevated risk of stroke as it shares anatomical, embryological, and physiological similarities with the brain [ 20 ] . Hence, retinal fundus imaging is considered a potential biomarker of CVDs as a complementary imaging technique for stroke diagnosis. Fundus images are acquired by a fundus camera, which is a portable and non-invasive device generally used by ophthalmologists [ 21 ] . Accordingly, retinal arterial abnormalities associated with cerebral infarction and hemorrhage are revealed with retinal fundus imaging [ 22 , 23 ] . These abnormalities include arteriolar fibrous thickening near the optic disc, retinal venular widening, increased tortuosity, arteriolar narrowing, and cotton wool spots [ 20 , 23 ] . Moreover, transient or permanent vision loss is a major outcome of stroke that can be caused by retinal ischemia [ 24 ] . Therefore, retinal imaging is strongly suggested as a diagnostic tool in the early prediction of stroke [ 22 ] . In healthcare, deep learning is extensively used over medical imaging data, where computer-aided diagnosis of various diseases and medical emergency conditions is performed achieving state-of-the-art results. Deep learning-based approaches with Convolutional Neural Networks (CNNs) are widely used in stroke imaging including applications such as prediction, treatment, decision making, and segmentation of lesion regions in the brain [ 25 ] . In the literature, many studies [ 26 , 27 , 22 , 28 , 29 , 30 , 31 ] have performed ischemic stroke detection using retinal fundus images with deep learning. Pachade et al. [ 26 ] compared the performance of hand-crafted and self-supervised deep learning features extracted from fundus images in stroke detection, and achieved an Area Under Curve (AUC) of 0.57 using the k-Nearest Neighbors classifier. Jeena et al. proposed a stroke detection scheme by extracting various hand-crafted features in their several studies [ 27 , 28 , 22 ] . In another study [ 29 ] , authors evaluated a state-of-the-art CNN model to predict ischemic stroke using fundus images. Moreover, transfer learning has been performed in studies [ 30 , 31 ] , where pre-trained deep networks have been trained further with retinal fundus images, yielding AUC scores of 0.67