A nonlinear multiresolution approach to echocardiographic image segmentation

The application of multiresolution nonlinear window operators to echocardiographic image segmentation is studied. A method that can potentially reduce the signal-to-noise ratio loss in the data is to digitize the line-mode data directly. An approach that is more stable than using finite difference operators is to compare and correlate different beams of the line-mode fan-beam data. Nonlinear operators such as median filtering and morphological operators are used. These operators are known to not displace the edges in the data while having robust performance in noise suppression. After applying an operator of a specific size, the data sequence can be segmented into different regions. By using such an operator with different sizes, the amount of detail can be reduced successively, to the point where only the major components, such as the inner heart chamber, remain and are identified. The endocardial boundary is thus detected based on the edge points found on each beam