Heart attack risk prediction using Retinal Images

The timely and precise diagnosis of heart disease is essential, as it ranks among the leading causes of mortality worldwide. Traditional techniques like ECG and angiography are effective; however, angiography is invasive, expensive, and often inaccessible in many rural and resource-limited areas. This study proposes a non-invasive approach to predict heart attack risk by analyzing retinal images, which provide critical insights into cardiovascular health due to the established link between retinal microvascular changes and heart conditions. Our study aims to categorize retinal fundus images into two categories: CVD and non-CVD. Analyzing retinal images alongside clinical data has increased prediction accuracy using advanced deep learning algorithms, such as VGG19, ResNet, DenseNet, InceptionV3, and MobileNet. The approach was mainly based on pre-processing for improving image quality, segmentation of vascular features, and training on annotated datasets. Experimental results show that clinical data-integrated retinal images classify with significantly better accuracy than models based solely on retinal images. It is a non-invasive and affordable method for scalable early detection of heart disease and could help save lives by increasing access to such tools.