Retinal imaging is often used to diagnose and monitor eye diseases, but much more can be revealed: abnormalities in the retinal microvascular range may indicate a wider range of vascular diseases, including heart problems.
Recently, an international team of interdisciplinary disciplines at the University of Leeds has developed an artificial intelligence system that can automatically analyze routine eye scans and identify people at risk of heart attack. Their findings were published Jan. 25 in the journal Nature Machine Intelligence.
The eye is the window of the mind: information about heart health can be predicted by scanning the eye | University of Leeds
Cardiovascular disease is one of the most important causes of premature death worldwide. Early identification and preventive treatment will help reduce its prevalence. Currently, a patient's risk of cardiovascular disease is estimated by age, sex, smoking status, family history, and medical imaging such as coronary CT, echocardiography, and cardiovascular MRI. However, this imaging is usually done in hospitals and is more expensive, limiting their potential for use in countries with fewer medical resources.
"This AI-based technology opens up revolutionary possibilities for heart disease screening," corresponding author Alex Frangi explained to the media, "Retinal scanning is relatively inexpensive, and many places with glasses will also be equipped." According to the results of AI screening, people at high risk of developing the disease can be referred to a cardiologist for treatment. This scanning technique could also be used to track the early signs of a heart attack. ”
Retinal images | UK Biobank
Lead author Andres Diaz-Pinto and his colleagues investigated whether the AI system could use retinal images to assess the volume and efficiency of pumping blood in the left ventricle of the heart. Ventricular enlargement is often associated with an increased risk of heart disease and can be used to predict the likelihood of a future heart attack.
To do this, the researchers trained a multi-channel variational autoencoder and a deep regression network to estimate the left ventricular end-diastolic volume (LVEDV) and left ventricular mass (LVM) directly from retinal images. They used 5663 and 71515 data from the UK Biobank Imaging Study to train and validate two networks of this AI system.
The first set of data includes cardiac MRI images, high-quality retinal images, and demographic data, while the latter have only high-quality retinal images and demographic data. Compared with the true depiction of MRI images, the researchers found that retinal images could be used to quantify the parameters of the heart.
Next, the researchers used LVM/LVEDV estimated by retinal images combined with demographic data, or demographic data alone, to predict whether patients were at risk of heart attack in the following 12 months. For comparison, they used data from the UK Biobank, where retinal images were not used to train AI. Of them, 992 suffered myocardial infarction after taking images, and 992 did not.
They found that adding LVM/LVEDV estimated by retina images to demographic data could improve the predictive performance of AI systems. The system predicts future myocardial infarction events through retinal images, with sensitivity and specificity of 0.74 and 0.72, respectively, which only requires adding two additional variables of age and sex (usually both ophthalmologists or opticians have this information).
Finally, the team conducted external validation using retinal images and demographic data from the independent database, the NIH Age-Related Eye Disease Studies, which included 180 cases of myocardial infarction and 2,830 cases without myocardial infarction.
The researchers found that the algorithm's ability to predict myocardial infarction was influenced by retinal images of older macular degeneration. The predictive performance of this AI system was highest in cases without macular degeneration, with sensitivity and specificity of 0.70 and 0.67, respectively. With the severity of macular degeneration, the predictive power continues to decline. The team notes that retinal diseases, such as macular degeneration, interfere with algorithms inferring systemic cycle features from retinal circulation features.
The researchers conclude that the AI system they developed can assess the health of the heart and predict the likelihood of myocardial infarction through inexpensive and readily available retinal photographs and demographic data. Such a system, they recommend being used in eye clinics and optical stores, to assess the likelihood of a heart attack in patients, recommending that they do further cardiovascular checkups.
bibliography
[1]https://physicsworld.com/a/routine-eye-scans-could-provide-cost-effective-screening-for-heart-disease/
Compilation: Small pot
Typography: Yin Ningliu