Reconstruction of echocardiographic images for the detection of several congenital heart defects

The purpose of this thesis was the research of a new segmentation method to detect and extract the heart cavities in echocardiographic images. To satisfy the clinical practice requirements, it is demanded that the segmentation algorithm is capable of providing reliable boundary extraction. This is crucial to the correct diagnosis of potential congenital malformations and diseases. The proposed algorithm is based on Log-Gabor wavelets to detect symmetric features in the images, and a level set evolution in order to extract simultaneously all heart cavities in an accurate way. The formulation of the level set uses a new logarithmic based stopping function, which improved the boundary detection when compared to other level set methods. Experiments were performed on echocardiographic images of children hearts. The validation of the algorithm included comparisons using stateof- art methods as the manual contours drawn by a trained physician, and the error quantification using similarity metrics. Our method outperforms the state-of-art in echocardiographic heart segmentation, encouraging its future application in the clinical practice. This new segmentation method has potential to improve the performance of 3D reconstruction algorithms, since the increased accuracy of the extracted heart contours simplifies its alignment in space and henceforth the recover of the 3D structure