During the past few decades, echocardiography has emerged as a low-cost standard imaging modality in cardiology because of its ability to provide non-invasive images of the heart. An echocardiogram, despite its image artifacts, provides detailed information about the structure and function of the heart. To enhance chamber segmentation accuracy in this complex organ, an atlas-based approach is utilized, leveraging prior knowledge for a fast, simple framework. This study proposes segmenting left heart chambers and extracting boundaries using an atlas constructed with a second-order polynomial nonlinear transformation followed by an affine transformation, with iterative reference image updates to avoid bias. Registration between the atlas and subject images involves three steps: affine, second-order polynomial, and non-rigid transformations. In order to improve the accuracy in systole phase images, the Radon transform is applied to a small area around the middle border of the two chambers. The average Dice coefficient accuracy is 0.902±0.047 for the left Ventricle and 0.915±0.026 for the left Atrium in systole, and 0.912±0.031 and 0.881±0.045 in the diastole phase, respectively. This method, applied to apical four-chamber views, also can be used in other echocardiographic views.

Keywords