Automatically Driven Vector for Guidewire Segmentation in 2D and Biplane Fluoroscopy

The segmentation of endovascular tools in fluoroscopy images can be accurately performed automatically or by minimum user intervention, using known modern techniques. It has been proven in literature, but no clinical implementation exists so far because the computational time requirements of such technology have not yet been met. A classical segmentation scheme is composed of edge enhancement filtering, line detection, and segmentation. A new method is presented that consists of a vector that propagates in the image to track an edge as it advances. The filtering is performed progressively in the projected path of the vector, whose orientation allows for oriented edge detection, and a minimal image area is globally filtered. Such an algorithm is rapidly computed and can be implemented in real-time applications. It was tested on medical fluoroscopy images from an endovascular cerebral intervention. Experiments showed that the 2D tracking was limited to guidewires without intersection crosspoints, while the 3D implementation was able to cope with such planar difficulties. Keywords—Edge detection, Line Enhancement, Segmentation, Fluoroscopy.