GENERATION AND SEGMENTATION OF 3D MODELS OF BONE FROM CT IMAGES BASED ON 3D POINT CLOUDS

Abstract

The creation of 3D models of bone from CT images has become popular for surgical planning, the design of implants, and educational purposes. Software is available to convert CT images into 3D models of bone, however, these can be expensive and technically taxing. The goal of this project was to create an open-source and easy-to-use methodology to create 3D models of bone and allow the user to interact with the model to extract desired regions. The method was first created in MATLAB and ported to Python. The CT images were imported into Python and the images were then binarized using a desired threshold determined by the user and based on Hounsfield Units (HU). A Canny edge detector was applied to the binarized images, this extracted the inner and outer surfaces of the bone. Edge points were assigned x, y, and z coordinates based on their pixel location, and the location of the slice in the stack of CT images to create a 3D point cloud. The application of a Delaunay tetrahedralization created a mesh object, the surface was extracted and saved as an STL file. An add-on in Blender was created to allow the user to select the CT images to import, set a threshold, create a 3D mesh model, draw an ROI on the model, and extract that region based on the desired thickness and create a new 3D object. The method was fully open-sourced so was inexpensive and was able to create models of a skull and allow the segmentation of portions of that mesh to create new objects. Future work needs to be conducted to improve the quality of the mesh, implement sampling to reduce the time to create the mesh, and add features that would benefit the end-user.