As I understand: HU = CTnumber * Slope + intercept.
The two values are slope and intercept, in turn, takes the tag (0028, 1053) and the tag (0028, 1052), using ITK.
When displayed, the pixel intensity is used: Pixel_intensity = ((HU-P1) * pow (2, i-1)) / W.
Value of HU is determined by each machine: CT, MRI, …
Algorithms for image processing (filters, segmentation) only affect pixel_intensity. HU will not change during image processing. Is that right?
In my problem, geometry is built from HU value through the standard conversion function between HU and density.
So, as I understand, the application of image processing algorithms also does not make sense or affect the construction of geometry?
Not really. The Hounsfield scale is a measure of radiodensity and is linked to the attenuation coefficent of tissues. That why it is so useful, because you can compare images independently of its origin (device, manufacturer, etc.) due to all CTs being calibrated on water.
When using DICOM, the pixel value are usually encoded as unsigned 16-bit integers in the PixelData buffer. To retrieve the true value (i.e. in HU, which is equivalent to “CTnumber” by the way), you apply the transform : HU = PixData*Slope + Intercept .
Given that the pixel value == HU, if your algorithm change these values, you cannot rely on the scale anymore to infer density.