Detecting cardiovascular diseases from radiographic images using deep learning techniques

Abstract

Cardiovascular disease (CD) is one of the leading causes of death and disability across the globe. Chest x-rays (CXR) are crucial in detecting chest and CD. The CXR images present helpful information to the radiologist to identify a disease at an earlier stage. Several convolutional neural network (CNN) models for classifying the CXR images have been established. However, there is a demand for significant improvement in CNN models to generalize them in diverse datasets. In addition, healthcare centers require an effective model for identifying CD with limited resources. Therefore, the authors developed a CNN-based CD detector using CXR images. The proposed research employs the You Only Look Once, version 7 technique to extract features and DenseNet-161 for classifying the CXR images into normal and abnormal classes. The authors utilized datasets, including CheXpert and VinDr-CXR, for the performance evaluation. The findings reveal that the proposed study achieves an accuracy and F1-measure of 97.9, 97.47, 96.85, and 97.77 for the CheXpert and VinDr-CXR datasets, respectively. The recommended model required fewer parameters of 5.2 M and less computation time for predicting CD. The study's outcome can assist clinicians in detecting CD at the earliest stage.