An End-To-End Deep Learning Approach for Atherosclerosis Detection using Autoencoders

Cardiovascular disease remains the primary cause of mortality worldwide. Early diagnosis of cardiovascular diseases, particularly atherosclerosis, has been the focus of extensive research. Gated Magnetic Resonance Imaging has enabled researchers to capture high-resolution images of the cardiovascular system, facilitating the detection of arterial wall plaque. The integration of machine learning techniques into the diagnostic process has significantly improved the accuracy of cardiovascular disease diagnosis. This paper proposes a deep learning approach that utilizes autoencoders to identify atherosclerosis in murine subjects from gated MRI images. The images underwent preprocessing using a Block Matching 3D (BM3D) filter to eliminate noise and Contrast Limited Adaptive Histogram Equalization (CLAHE) to enhance contrast. To evaluate the performance of the proposed model, we trained and tested two different autoencoder architectures, stacked and convolutional. The models successfully distinguished healthy and diseased subjects based on input and reconstructed images. The Convolutional Autoencoder outperformed the Stacked Autoencoder by minimizing the reconstruction loss by 95%, demonstrating superior accuracy in reconstructing medical images.