YOLO and Morphing-based Method for 3D Individualised Bone Model Creation

The paper presents an automated tool for individualised skeletal model creation and identification of specific bone structures from computed tomography (CT) images. Contrary to the classical segmentation methods, the presented one is aimed at exact recognition of the bone joints. It takes advantage of a reference skeletal model. The developed procedure relies on the neural network detection of the bounding boxes to select single bones from CT slices. In the subsequent steps the reference model elements are matched to the detections by means of affine transformation and next, by morphing algorithm. The method incorporates various image processing algorithms including swarm intelligence heuristic algorithm-Artificial Ant Colony (enriched with an original pheromone deposition function). Three-dimensional models of bones created by the procedure can be used in the task of automatic delineation of bone structures, positioning of the patient for radiotherapy or in planning of orthopaedic medical procedure. They can also serve as a visual help in diseases detection.