Abstract: Cardiovascular diseases (CVDs) appear to rank highest in the global mortality rate and thus early diagnosis of these diseases is important gesture to be observed. Based on that, the current work will endeavor to develop a CNN model with MobileNetV3 for screening of CVDs from the retinal images. As for the specific details on the model to use, MobileNetV3 is selected because it is demonstrated to impart higher performance with less computing burden, and CNN layers to extract prominent features from the images. Hence for enhancing the quality of the retinal images of the given dataset which consists of multiethnic population data and includes two groups; with and without CVDs the images are subjected to resizing, normalization and image augmentation. The features incorporated in the architecture of the CNN designed for the prediction of the images of the retinas are such that the images of the retinas with and without CVDs can easily be distinguished. That way, the model could maintain reasonable degrees of reliability particularly in categorizing, analyzing and cheap diagnosing the cardiovascular diseases. The custom CNN achieved a test accuracy of 79.69%, while the MobileNet-based model demonstrated superior performance with a test accuracy of 90.23%. These results indicate the potential of deep learning, particularly transfer learning, for developing efficient and accurate tools to aid in the early detection of retinal pathologies, potentially improving patient outcomes and accessibility to eye care.

Keywords: Retinal images, deep learning, Convolutional Neural Network, MobileNetV3, cardiovascular diseases (CVDs), early diagnosis, medical imaging, health care, biomarker- based non-invasive diagnostic, image classification.

| DOI: 10.17148/IARJSET.2025.125341