Digital Image Formation
Different photon ranges used in X-ray have different applications, ranging from X-ray crystallography, mammography, medical CT or airport security. X-rays are used in 2D or 3D, like in CT.
What is a digital image?
- n-dimensional matrix of intensity values
- each cell of matrix is pixel (2D) or voxel (3D).
- but also 4D: collection of 3D images over time.
Random facts:
- Healthy lung tissue is dark
- Pixel values of X-rays are in ... Enhancement Unit Values? (Didn't hear this clearly)
From analogue to digital
In reality we have continuous values in space and in intensity.
Two steps:
- Sampling
- Quantization
Sampling
Ideal sampling s(x,y) in regular grid of M x N locations is a collection of dirac functions $\delta{x,y}$ that we shift over that grid.
$$s(x,y) = \sum_{j=1}^M \sum_{k=1}^N( \delta( x-j \Delta x, y-k \Delta y ))$$
Note that $\Delta x$ and $\Delta y$ can be different. So the sampling is not necessarily isotropic.
For the transformation from world coordinates to pixel/voxel coordinates we need to know the origin of the input image, which is determined by the particular type of device i.e. it's manufacturer.
Sometimes you need to know how the real image was digitalized.
You can do this e.g. with Itk_img.GetSpacing().
For example, if you have two consecutive pixels, you often want to know whether they cover let's say 1mm or 1cm (again: may be different in x and y direction).
Quantization
Then for each pixel, we need to map intensity values to something that can be stored in bits.
The quantization is determined by your bit depth. If you only have 1 bit, you can only do binary thresholding with two color values. A very common quantization is 8-bit, where you can represent 256 values [0,255].
In X-ray CT intensity is measured in Hounsfield Units (HU) in range [-1000,3000], where -1000 is air, -400 - -600 is lungs, 1000 is bones and 3000 is very dense bone. So there is a close correspondence between these values and body parts.
This is very important information, because this allows us to determine appropriate thresholds for identifying particular body parts.