Human heart is considered one of the most import organs of the human body, since its job is to provide the body with blood. One of the methods that clinicians utilize, to examine the heart and its internal structure condition, is the TransThoracic Echocardiogram (TTE), which is the most used, agile, and cost-effective cardiac imaging modality. Machine Learning techniques and Deep learning neural networks, implemented in TTE images, can deliver highly accurate and automated interpretation of heart’s clinical condition, which can greatly assist cardiologists in their evaluation of heart’s abnormality or not. In the current master thesis, a deep learning algorithm will be examined in various dataset resolutions and a comparison of its performance on the task of classification of the enlargement of the left atrium of the human heart, with the use of TTE images from patients of a Greek Hospital, will be studied. The basic algorithm is a combination of a Unet and a Convolutional Neural Network (CNN). Unet will segment the A4C TTE images over the cardiac Left atria (LA) and CNN will classify the segmented images for normal or abnormal size of the LA. Addittionaly a Semi-supervised GAN will be trained and evaluated in classifying the cardiac LA as normal or abnormal.