Improving speed and clarity in x-rays
Clear x-ray images are crucial for medical imaging and especially during radiation therapy. Pinpointing the exact location of a tumour helps to target treatment whilst protecting healthy tissue.
Any respiratory or muscular movement in a patient during x-ray can cause blurring in radiographic images. Known as motion artifact, this blurring is a common issue in medical imaging. We worked with CERN to develop the Tomographic Iterative GPU-based Reconstruction toolbox (TIGRE) to address this challenge.
Modifying algorithms and translating techniques
The toolbox is based on Cone Beam Computed Tomography (CBCT). This is a scanning process that takes a series of 2D X-ray pictures and processes them into a 3D image.
We reviewed a range of published CBCT algorithms and modified them to run on a laptop fitted with a graphic processing unit (GPU). Using the same sort of graphic processor you find in games consoles, we were able to adapt the algorithms to be faster.
A second part of our research explored motion correction. We built on phase space tomography techniques developed 20 years ago by Dr Steven Hancock at CERN. Phase space tomography uses known motion models to update tomographical information during algorithmic image reconstruction. This removes all movement occurring in the image. Our research translated this technique for use in x-ray tomography.
The toolbox uses advanced iterative and regularised reconstruction methods. It can work with a wide variety of iterative algorithms, including:
- Feldkamp-Davis-Kress (FDK) algorithms
- simultaneous algebraic reconstruction technique (SART) algorithms
- Krylov subspace algorithms
It has GPU-accelerated projection and back projection and covers a range of methods using total variation (TV) regularisation. Its modular design makes it easy to add in new algorithms.
Bringing research benefits to communities
TIGRE offers a simple and affordable way to improve imaging and reduce radiation doses for patients. The accelerated algorithms in the toolbox make medical imaging processing run around 1000 times faster. It is now fast enough to be used in clinical scenarios.
These algorithms can improve image quality and in some cases work with very low amounts of data. They have the potential to reduce radiation doses in patients and increase survival rates.
The software is released under an open source licence, making it available to everyone. It is hoped that this will encourage hospitals, industry and wider research communities to benefit from our research.