Automatic Fiducial Detection is a tool for open4Dnav, which is developed with open-source libraries. It detects, localizes three dimensional (3D) objects (titanium balls) automatically in preoperative radiological images for given geometric object properties using morphological algorithms iteratively and allows geometric measurements, or computation operations by detected objects using a special filter.

Goal
The goal of the project is to develop an algorithm to identify/detect in the patient inserted and in the near of the surgical area positioned titanium fiducials in computed tomography (CT) images automatically, to calculate their centroids, locations, and relationships to each other, in order to realize the registration of CT images with the patient for the surgical navigation in preoperative phase without any user error.

Automatic Fiducial Detection Tool enables detection of searched objects in the background in seven steps. First is defined using the user an unique working area “region of interest (ROI)” in 3D multiplane view images that contains desired objects in it and thus by ROI reduce the processing time of the algorithm. In the second step, a 3D binary image is generated by using of the binary threshold method, which is used later for the detection, localization and measurements of geometric features of labeled objects. After segmentation or rather binarization, a structured 3D binary ball element is created based on the radius of searched objects that used in the next step in the morphological process. In the fourth step, a morphological opening filter on the binary image is applied with the previously created element in order to determine whether the object in binary image corresponds to the predefined object element. After the elimination of unrequested objects (faults) in the image, a geometry filter on the output image is applied, that is generated after morphological step, to determine and calculate the geometrical properties (centroid-coordinates in cartesian coordinate system, major/minor axes lengths, number of detected objects, elongation, eccentricity, etc.) of detected objects. In the sixth step, a distance map based on found geometric properties is drawn, that represents and simulates the relations of the localized objects, e.g. distances between objects. At the end of the process the detected objects are segmented and visualized as an iso-surface or rather 3D image.

Screenshots

Detailed information can be found in the Downloads and Publications pages.

Y. Özbek, W. Freysinger

• 
•