View Full Version : Summary: Correlation the HU of CT images to bone density

María Prado Nóvoa
04-21-2003, 11:20 PM
Dear colleagues,

This is the summary of responses to my query regarding "Correlation the HU of CT images to bone density", the original query followed by the responses

Thanks to everyone, Your responses were very helpful.


María Prado



************************************************** *********

Dear Colleagues,

I am trying to identify the mechanical properties of tibias of deficient knees (OA, TKA, etc). I want to relate their mechanical properties to the wet apparent density of bone, as it's reported in the literature.

For this purpose I am using standard CT data, but I need to correlate the HU of the CT images to the apparent density, therefore I need to calibrate the scanner, specifically for the range of the trabecular bone of the human tibia if it is possible.

Can anybody let me know how can I approach this task?

I know that a bone equivalent calibration phantom is needed, does anybody know how is this phantom? I believe that I can use water and a piece of wet bone; do you know if it is ok? Is it significant where I put the water (glass, plastic.)?

I would appreciate very much any comment about this questions or any other information on the calibration of CT images.

I will post a summary of responses if there is interest.

Thanks for your help,

María Prado
University of Málaga.
E-mail: mpn@uma.es

************************************************** *********


There are equations out there that correlate DRY apparent density to HU, but I haven't seen any equations for WET apparent density.

You should look at a study done by JY Rho et al entitled "Examination of several techniques for predicting trabecular elastic modulus and ultimate strength in the human lumbar spine." (Clinical Biomechanics, 1994. Vol 9:67-71). The authors describe their calibration techniques in their methods section (relating water and air to HU). I hope this helps. If you don't have access to the paper, please let me know.

I sent out a question (about the conversion from HU to density) almost two months ago and I got no response. My question was related to the conversion from HU to density measured by DXA (g/cm^2).

Please forward to me any responses you get regarding your question.

Good luck,


************************************************** *********

Dear María Prado,

You should get in contact with PD Dr. Felix Eckstein, LMU München, Anatomische Anstalt (mailto:eckstein@anat.med.uni-muenchen.de) He and his group have published quite a lot on bone density. Yours sincerely,

Erich Brenner

************************************************** *********


> For this purpose I am using standard CT data, but I need to

> correlate the HU of the CT images to the apparent density,

> therefore I need to calibrate the scanner, specifically for the

> range of the trabecular bone of the human tibia if it is possible.

> Can anybody let me know how can I approach this task?

As defined in some literatur (i.e. schultz, Computer- Tomographieverfahren, Thieme, Stuttgart - its in german) the HU is defined for quantitative CT as follows:


where t=11, rwater coefficient of water and rair is the coefficient of dry air. You do not need to calibrate your CT for HUs, but only measure dry air and water. After this you can simply calculate HUs from CT data, if these contain attenuation coefficients (quantitative CT). Hope this helps.


Andreas Boehm

************************************************** *********


Im not exactly sure if this is what you are asking, but I remember finding

out the following before:

A Hounsfield Unit, in CT scanning describes the amount of X-ray attenuation of each voxel (volume element). These voxels are each given a numerical value which represents the radio-opacity of the voxel, ranging normally from -1024 HU to 3071 HU (2 to power of 12 values).

These are calibrated so that -1024 HU is the value given to air (1kg/m3) and HU is given to water (1000kg/m3). The relationship between HU and density is also linear (I read that in a number of places). Using that info. you can switch between density and HU for any value. eg. A voxel whose HU equals 200, will have a density = 1223 ie. density = 1+ (1000-1) (1224/1024) ie. density = 1+ (1000-1)[(200 - (-1024)]/[(0 - (-1024)] You should expect to find the HU of bone to be in the region of 1000 (cortical bone) down to 200 in the centre. (If this is 2000 to 1200, there may simply be a shift in the scale of the HU values, such that AIR takes a HU value of 0, instead of -1024)... In this light, I would recommend using air and water as your two calibration points and use a linear relationship between them.

Hope this helps, Niall

PS: I'd be very interested to hear other inputs

Hi Maria,

We have done some similar research where we had to calibrate CT scan data. We used different concentrations of dipotassium hydrogen phosphate solution in test tubes (4%, 8%, 16%, 32% and 64%). For more information see our publicationbelow:

Mootanah R, Van Der Linde I, Ingle P, Cheah K, Dowell J and Shelton JC. An accurate three-dimensional finite element model of the pelvic bone with geometry and material properties retrieved from CT-scan data. Computer Simulations in Biomechanics, 8: 81-84, July 2001, Libreria Clup.

Rajshree Mootanah, PhD
Bioengineering Research Group
Design and Engineering - DaCS
Anglia Polytechnic University (APU)
Bishop Hall Lane
Tel: (44) 1245 493131 ext 3316
Fax: (44) 1245 252646

************************************************** *********

Dear colleague,

CT provides a measurement of linear attenuation of tissues. In particular the linear attenuation µt of each tissue pixel is compared with that µw of water by the formula:

CT number = 1000 (µt -µw) / µw

(also different coefficient can be used instead of 1000)

Water is used as a the reference material because its attenuation coefficient is close to those of soft tissues and it is a reproducible material for machine calibration. The CT number (or Hounsfield number) for tissues depends on the kV employed. For example at 80 keV the linear attenuation coefficients of water is 0.19 cm-1 and 0.38 cm-1 or higher for cortical bone, with a resulting bone CT number of more than 1000. Linear attenuation at a defined energy is mainly dominated by the atomic number Z which is closely related to tissue apparent density. We can say with good approximation that the tissue apparant density is the driving parameter of the attenuation coefficient with a direct linear correlation. Therefore the higher is the apparent density of the tissue and the higher will be the measured CT number.

A specific calibration problem can arise from the question of the evaluation of the bone mineral density of bone segments, as provided by Quantitative Computed Tomography (QCT) systems. In QCT systems the calibration is performed with a reference phantom scanned with the patient. An internal calibration system is normally available as a tool of each CT device, but external phantoms are also available. The phantom is usually manufactured by use of calcium hydroxyapatite (HA), the mineral component of bone and plastic water equivalent to simulate soft tissues. For example we had a good experience with the European Spine Phantom (ESP)

Eur J Radiol 1995 Jul;20(2):83-92

The European Spine Phantom--a tool for standardization and quality control in spinal bone mineral measurements by DXA and QCT.

Kalender WA, Felsenberg D, Genant HK, Fischer M, Dequeker J, Reeve J.

By this phantom the measured CT Hounsfield units can be converted in terms of mg (HA) / cm3 of bone tissue. Mineral density is a very useful parameter because it is able to explain most of the mechanical properties of bone. Moreover the direct calibration obtained with HA as a reference can give us density measurements less sensitive to the x-ray beam characteristics over time. Unfortunately, different and not standardised solutions have been adopted by CT manufacturers, causing the energy and reference material dependency of QCT mineral measurements from different devices.

The relation between Hounsfield units vs mineral density is almost always linear. An example:

Device: QCT GE Syec 300
Calibration: default calibration method provided by GE


hounsfield units = 2.783 + 1278 * BMD

Bone Mineral Density (BMD) has to expressed in g/cm3

R2 of regression = 0.998

Do not forget that the regression law CT Hounsfield Vs BMD depends:

-on the specific QCT device

-on the specific phantom and calibration protocol adopted

Therefore, let me suggest you to abandon any idea to develop your own phantom...

Hope it helps. Ciao, Fabio Baruffaldi

María Prado
Department of Mechanical Engineering.
University of Málaga.
Campus El Ejido s/n.
29013 Málaga. SPAIN.
Phone: +34 952 131314
Fax: +34 952 13269
E-mail: mpn@uma.es

To unsubscribe send SIGNOFF BIOMCH-L to LISTSERV@nic.surfnet.nl
For information and archives: http://isb.ri.ccf.org/biomch-l