An Effective Deep Learning Approach to Cardiovascular Disease Detection Using Retinal Fundus Images

Cardiovascular diseases (CVDs) are recognized as a significant global health challenge, contributing to an alarmingly high mortality rate worldwide. Each year, millions of lives are lost, and billions of dollars are spent in the healthcare sector due to CVDs. Addressing the global burden of CVDs requires advancements in accurate detection, which plays a crucial role in saving lives and preventing severe health complications. Deep Learning (DL) and Machine Learning (ML) approaches present substantial advantages for detecting CVDs. In this paper, CVD detection is performed by leveraging DL and ML-based methods. A set of Convolutional Neural Network (CNN)-based pre-trained DL models, including DenseNet121, DenseNet169, DenseNet201, VGG16, VGG19, Xception, ResNet50, ResNet101, and ResNet152, were used as feature extractors. A custom DL-based classifier was designed, and several ML models, including Support Vector Machine (SVM), Random Forest (RF), and Extreme Gradient Boosting (XGBoost), were employed as classifiers. Retinal fundus images extracted from 6 datasets were used in this study. The hybrid model combining DenseNet169 and Xception achieved the highest accuracy (93.37%), precision (97.70%), F1-score (92.90%), and AUC (97.83%) with the custom classifier, while the Random Forest (RF) classifier recorded the best recall at 92.71%. This performance is exceptionally high and ranks among the top results reported.