Segmenting skull bone in medical images (head CT scans of the patients).
We can use two approaches for the task. The first one is Traditional Methods and the second one is Deep Learning.
These image processing methods basically separate out different parts of the images based on the intensity values of the pixels. Some of the basic operations used are.
- Thresholding
- Windowing
- Morphology (Erosion, Dilation, Opening, etc.)
- Blurring
- Masking and so on.
Deep learning-based methods require annotated datasets with ground-truth masks. We can use these maasks along with image datasets to train a deep neural network like UNet architecture. Such networks perform betetr than classical approaches nowadays and are fully automated . However, deep learning-based methods usually require large datasets for training purpose.
For now, to segment skull bone, we only focus on traditional methods. The reason for that is that bone intensity is usually very high compared to other parts of the image, so we can easily segment it using techniques like thresholding and other methods. However, we can still use deep learning for this task and make it fully automated. For now, that is in future goals and we will only use classical methods at the moment.
I used the following main libraries.
- Pydicom
- Scikit-Image
- OpenCV
We need pydicom to read the DICOM files. If your data is in Nifti (.nii) format you can use nibabel library, or you can also convert your data using 3D Slicer.
Note that you can use either OpenCV or Skimage for image processing or you can also combine operations from both libraries. Usually, one library is enough to accomplish a particular task. OpenCV is maybe a more common and go-to choice for many people. However, I found some more functions in Scikit Image and wanted to check them as well. It comes down to your choice in the end.
Steps are pretty straight forward as you can see in the notebooks. We just load the image and apply different operations like thresholding and morphology. In the end, we get extracted mask of our skull.