Executive Summary : | Echocardiography is a widely used cardiac imaging test for identifying cardiac diseases, and its advantages include its affordability, superficiality, availability, and comfort for patients. Accurate echocardiography (echo) images are crucial for evaluating the heart's functionality, but ultrasound images can be corrupted due to speckle noise. Despeckling is essential for recovering images and identifying edges more prominently. A wavelet-based adaptive thresholding technique is proposed to despeckle echo images. Segmentation of echocardiography images is crucial for estimating cardiac parameters like heart wall thicknesses, ejection fraction, end diastole, and end systole volume. The left ventricle (LV) plays a vital role in pumping oxygenated blood throughout the body. Traditional methods involve manual segmentation, which is time-consuming and differs among individuals. Deep learning algorithms using Convolutional Neural Network (CNN) models have gained significant applications in image investigation, including classification, segmentation, and object identification. The proposed research proposal focuses on automatic segmentation of the left ventricle (LV) from echo images using a CNN. Preprocessing, including denoising and feature extraction algorithms, is incorporated with the deep learning-based CNN model to improve prediction accuracy. The proposed system needs to be trained on a large dataset of echo images, including thousands of echo images of the left ventricle. The CAMUS dataset, which includes 450 patients' data for training purposes and 50 patient datasets for testing, will be used for testing. |