An approach to automatic segmentation of 3D intravascular ultrasound images

Intravascular ultrasound imaging provides high-resolution images of sections of the arterial wall. Artefacts (e.g. speckle) of ultrasound images and overlap of grey levels between different objects, however, often cause difficulties in automatic segmentation. Here, a method is proposed to reduce these difficulties. Firstly, the correlation coefficients between two grey levels is defined, by which the grey levels of an image can be clustered into different classes. Each of the grey level classes may represent an object. The pixels of the image are then classified using threshold technique. Pixels whose grey levels overlap in more than one class are misclassified at first and then corrected by testing their spatial relationship with their neighbours. Finally, isolated pixels are substituted by their neighbours. Experiments with real 3D intravascular ultrasound images have shown that this method is stable, fast and accurate in the sense of preserving the original image information. Extension of this method to segmentation of CT images also gives satisfactory results