Segmentation of Heart Valves and Walls From Ultrasound Images

Description:

        The purpose of this project is to isolate the heart structures from ultrasound images and to identify each structure based on its motion. The borders between the internal heart chambers and the structures around them may be determined in static images using simple gradient based methods or by applying active contour models to the images. The motion of these borders can then be tracked using optical flow methods. Training images can be used to characterize the motion of the various structures. The parameters derived form the training sets can then be used to classify endocardial borders as belonging to valve, wall, papillary muscle, or septum. In this way, these features may be extracted from the image based on their motion. The use of motion in feature extraction is advantageous in this case because it is often difficult to recognize features in static ultrasound images (see image below).

Data Source:

         The data used for this project is a fifteen frame sequence showing a cross-section of a beating heart, acquired at a rate of 30 frames/second. The data was aquired by gating the image aquisition to the EKG of a beating heart, so that a cross-section could be acquired at the same instance over a period of several heart cycles. The images are then averaged to give the final cross-section. Therefore, the sequence represents an average heart beat of the patient, as oppposed to one isolated beat.

        The data used has been extracted from a three-dimensional set of ultrasound data showing one 'average' heartbeat. Various coss-sectional views can be extracted from this data set, thus supplying the different sequences that will be used as training and as testing sets.

Typical image showing left and right
right venticals and the structures
asociated with them.

Goals:

• To extract the endocardial borders in 2-D images.
• To parameterize the motion of these borders.
• To associate the parameters with specific structures.
• To use these parameters to extract features in unknown images.