Paper Title
Accurate Cardiac Structure Segmentation from Echocardiographic Images using a Deep Neural Network: A Comprehensive Evaluation

Abstract
Accurate delineation of the left ventricle (LV) in 2D-echocardiographic images is essential for quantitative assessment of cardiac function, including ejection fraction, volume estimation, and wall motion analysis. Manual delineation is time consuming and suffers due to inter-observer variability, motivating the use for reliable automated approaches. This study presents an deep learning framework using Attention U-NE for automated LV segmentation from two-dimensional echo-cardiographic images. The model leverages attention gates to focus on anatomically relevant regions, thereby enhancing boundary delineation and suppressing background noise. The publicly available CAMUS dataset, which provides expert-annotated frames including diastolic and systolic, is used for training and evaluation. Preprocessing steps includes noise removal and contrast stretching to enhance quality of image, while extensive augmentation is implemented to improve model generalization. Experimental results prove that the proposed methodology exhibits better performance when compared to conventional U-Net models, achieving high Dice similarity coefficients (DSC) and low boundary errors, highlighting its potential for clinical integration in automated cardiac function analysis. Also model was evaluated based on Intersection Over Union (IoU), and Hausdorff Distance (HD). Keywords - Left Ventricle (LV), Echocardiographic Images, Attention U-Net, Deep Learning, Dice Similarity Coefficients.