Hello and welcome to my Honours project for Curtin University Bachelor of Radiation Science (Medical Imaging). My name is Katie Harris. Download this project by clicking under 'releases' on the right hand side of the main page.
This repository contains a set of tools for manipulating CT data to create digitally reconstructed radiographs (DRR) with Python. It was written in Jupyter Notebook using Anaconda Navigator to manage dependencies.
This Notebook can import a DICOM directory as a 3D numpy array, and then creates an average intensity projection over the entire dataset. The result closely resembles an x-ray but does not account for geometric factors such as beam divergence or source image distance. The Notebook contains some examples where the performance of this algorithm is timed, and some examples to demonstrate the same process in the sagittal and coronal plane.
These are a set of tools that let you create a simple DRR with perspective. It will let you import a DICOM directory, set geometric factors, and also save as a DICOM file. The DICOM saving function, save_dicom() is barebones. It saves to the same directory as the algorithm itself. It uses a default file from PyDicom and changes the image attributes. This means that the header information is not correct which protects patient information. The DICOM file will likely need windowing to look as expected once opened with any DICOM reading software.
projection_example is a Notebook that has the whole process working. projection_library is the functions alone, intended to be easy to import into other projects. Please note that you will want to run small-scale dummy data (eg. np.ones((3,3,3)) ) through both project() and add_img() to gain the benefit of Numba acceleration. Without acceleration, both functions will run very slowly.
This is a directory of material data based on publications from the National Institue of Standards and Technology (NIST). It is necessary for some functions to work. Save it to the same folder as the Notebooks. New files can be added as long as they match the format of existing files. Just be aware of adding entries to the 'CT numbers folder'; the CT numbers of the new material need to be sampled at the same interval as the other materials in that folder.
This library has the tools to import a DICOM as a 3D numpy array, and then models what the same scan taken at a different tube voltage would look like. It requires the material_info directory to work. It can account for a polychromatic x-ray beam if you have the anode material and aluminium filter thickness.kVp_shift_example.ipynb is a Notebook with this code compiling. The example Notebook provided demonstrates how this looks.
This set of tools is for importing a DICOM and creating a probable map of the mass densities. The purpose of this is to create a reasonable proxy of electron density. It requires material_info to work.
This is a set of tools for modeling Compton scattering. The rationale was that a change in direction of the photon produces a change in the position that the photon strikes the detector. If you know the origin of the scatter, and the probability distribution of the angle of the scatter (via Klein-Nishina equation), you can calculate the distribution of change in displacement. This is done with a convolution kernel. The convolution kernel is different depending on the area of the model because the angle of incidence of the the photon changes with divergence of the beam. The output is intended to represent scatter when projected using the projection library. This library has function to take a 3D numpy array and divide it into segments, convolve the segments in the xy plane then reassemble the model. This allows for modeling the effect of the diverging beam on scatter pattern, but produces a grid-like artifact at the edges of the segments.