Detecting endocardial boundaries by texture in two-dimensional echocardiography

An algorithm based on texture analysis for the extraction of endocardial borders in bi-dimensional echocardiographic images is presented. Using the Inverse Difference Moment (IDM) of the original image, an image that contains texture information is constructed and is used as a texture channel. Application of appropriate thresholding on projections of the cooccurrence matrix of the texture channel allows separation of regions and borders. The method has been applied to apical four chamber views and paraesternal short axis views, showing excellent results in determining reliable borders. Aditionally, by adding some detection criteria and medical knowledge of the heart, the algorithm can detect borders in incomplete and noisy images.