Novel Toolbox Generates CT Images from MR for Transcranial Ultrasound Stimulation Studies
A novel toolbox has been created to generate computed tomography (CT) images from T1-weighted magnetic resonance (MR) images for use in planning transcranial ultrasound stimulation (TUS) studies in humans.
This validated open-source toolbox allows for the creation of bespoke skull models for acoustic simulations in TUS studies using existing T1-weighted MR images for neuronavigation. The toolbox uses a 3D residual U-Net, pre-trained on 110 subjects and refined on a separate dataset of 37, to synthesize a 100 keV pseudo-CT image from a T1-weighted MR image. This toolbox fills the gap in existing methods and helps promote best practices and reproducibility in TUS. The toolbox is implemented in MONAI, an open-source deep-learning framework for medical imaging.
Key Takeaways
- The toolbox generates computed tomography (CT) images from T1-weighted magnetic resonance (MR) images for transcranial ultrasound stimulation (TUS) studies.
- It is aimed at researchers to generate individualized skull models for acoustic simulations in TUS studies.
- CT images are considered the gold standard for skull imaging, but obtaining CT images can be difficult due to radiation exposure and limited access to CT scanners.
- The toolbox uses a pre-trained 3D residual U-Net to synthesize a pseudo-CT image from T1-weighted MR image.
- The network was trained and validated on multiple datasets to increase its generalizability.
- The toolbox and accompanying code for transcranial simulations are open-source and have already been used by several TUS researchers worldwide.
- The toolbox is designed to be useful to both novice and expert users and can be used for transfer learning applications on their own datasets.
