A Multiscale Medial Axis Detection in X-rays Angiographies

In the present work we are concerned with the assistance to the diagnostic of coronaries stenosis from X-rays angiography modality. We are treating here the problem of the 2D multiscale medial axis segmentation. Using differential geometry we can characterize locally the medial axis of arteries as the image surface bottom lines of valleys. As Lindeberg , we use a normalized scale-space based approach to regularize the calculation of derivatives. Then, we can easily compute a local measure of valley for each scale of the process. The detection of the medial axis position is made searching the local maxima of the measure of valley in the direction of the first principal curvature. Then the medial axis and valley measure scale-spaces are used to obtain the 2D mutliscale medial axis of the coronary arteries. Then, the medical practitioner can choose the arteries of most interest for the future 3D reconstruction. We have tested our process on several angiographies. The medial axis is well extracted even in the case of strong stenosis. Furthermore, using the whole scale-space we limit the discontinuities of the medial axis detection. And, the computation cost is quite low.