Segmentation of echocardiographic images using classification in the radiofrequency feature space

We have developed a method for the digital acquisition of the radiofrequency (RF) data in echocardiography. We are able to acquire the complete signal from at least one (and the same) heart cycle. This approach enables us to study segmentation of the left ventricle based on classification of features of this RF signal. We have extracted and investigated several signal properties. These can be separated into two groups, first some traditional features used for segmentation (integrated backscatter, image moments, ...), secondly, properties that are not commonly used for segmentation purposes: features based on modeling the reflected signal using a "transfer function" approach. Once the parameters of the RF signal are extracted, the actual segmentation can be performed either using "simple" methods as thresholding or more sophisticated approaches like clustering in an n-dimensional feature space.