Background

The field of the invention is bone densitometry. Bone densitometry is used mainly to diagnose and monitor osteoporosis. The standard method is Dual-energy X-ray Absorptiometry (DXA), which records two projection X-ray images of different energies. By subtracting the two images the method determines the mass of Calcium per projected area, called the bone mineral density (BMD) .

BMD has been shown to be predictive of the risk of fractures in women .

One has also attempted to use BMD in children, but the size variation in a pediatric population has lead to problems interpreting the results, see [2] for a review. Since BMD is mass normalised to the projected area, it is sometimes called areal BMD.

An alternative method proposed for children is the use of pQCT which is a CT scanning typically of a part of the radius. This allows the determination of the true three-dimensional bone distribution and for instance the mineral weight per outer volume of a bone part. This is currently the most accepted method [3] .

Summary of the Invention

The invention addresses the need for a method of estimating the amount of bone in children. The desirable properties of the method are

• The method should be effective and easily accessible

• The radiation dose should be small

• The method should be well accepted by the child

• The method should be highly sensitive to deviation from normality

• The method should be able to cope with the large variation in body dimensions seen in a pediatric population.

The invention utilises plain projection radiography. This can either be traditional film or one of the new digital counterparts, like CR or DR systems. Thus the first step in the method is to

obtain a digital X-rays image, either by optical scanning of an X- ray film, or by fetching the image file from the digital X-ray system.

The use of radiography is an effective and easily accessible method because the equipment is part of general radiology. The measurement procedure is very fast, and the exposure itself is done in a fraction of a second, so that movement artefacts are a very minor problem.

The radiation dose is minimal, especially compared to pQCT. The method is well accepted by the child in contrast to pQCT - and to some extend DXA - where the child is immobilised in 1-10 minutes to allow for the scanning.

This accounts for the first three properties in the above list of desirable properties. In the following the invention is further described and it is explained that is has also the last two properties .

The invention extracts information from the image using radiogrammetry, i.e. measuring distances in the image. This is done automatically using a computer. A detailed description of how radiogrammetry can be automated is given in [1] . To locate the bones ono can use active appearance or active shape models or other standard techniques from image analysis. For a practical description of how this can be done, see [4],

Radiographs have excellent spatial resolution and it is possible to determine the outline of the bone very accurately. Also the inner boundary of the cortical shell in the shaft can be determined accurately. These boundaries are illustrated in figure 1, which shows an actual projection X-ray image with the found outer boundaries superimposed as a dotted curve. The cortical shell boundaries are shown with dotted lines with denser dots.

Radiogrammetry has been used in various guises for more than 40 years to quantify the amount of bone. The classical method is the metacarpal index defined as the cortical thickness T divided by the bone width W, both measured in the shaft of a long bone, typically the second metacarpal. This is illustrated in figure 2, which on the left shows a stylised projection image of a bone, where the bone length L, the bone width W, and the cortical thickness T are

indicated. On the right is shown a cross section of the bone and the transversal cortical bone area A is cross-hatched.

Another more recent method is the DXR-BMD method described in [1] . Here one forms the expressions DXR-BMD = c T or DXR-BMD = c T * (1 - T/W) where c is a constant determined so that DXR-BMD becomes an estimate of DXA-BMD in the radius based on T and W measurements in metacarpal 2-4.

The invention involves a new combination of information extracted from radiographs, designed to address the last two items in the list of desired properties of the method. To discover the proper combination, a clinical study was performed involving 629 children aged 7-17 years representative of the normal population. For each child the gender and age are known, and the X-ray images were analysed with active appearance models and radiogrammetry.

From the cortical thickness T, the bone width W and the bone length L one is free to form a measure of bone mass, which is most relevant for children. The starting point for the analysis is the transversal area A of the cortical bone:

A = π/4 * (W ² - (W - 2T) ² ) = π T W (1 - T/W)

This is computed under the assumption that the bone is cylindrically symmetric, but the expression is used also for non- symmetric bones. From this basic quantity six candidate measures were constructed:

A, A/L, A/W, A/L ² , A/ (LW) and A/W ² .

The candidate A/W ² = π * T/W * (1 - T/W) is approximately proportional to the metacarpal index 2T/W. A/W ² can be interpreted as the volumetric bone density, i.e. the bone mass per outer bone volume, and given the success of pQCT mentioned above, it is a promising candidate.

The candidate A/W is the cortical bone volume per projected area and is proportional to the above mentioned method DXR-BMD. The design criterion for choosing among the candidates was the principle that the measure should have minimal population standard deviation relative to its mean values in each age and gender bin. This derives from the conjecture that for normal subjects, the body

balances the amount of bone to the size of the body and the development stage, so that there is neither too little nor too much. The measure with smallest relative standard deviation is the measure being balanced, and its mean value is the constant of nature describing normality at that stage. This measure is hence the most relevant for diagnosis of disorders that disturb the optimal bone balance. With a standard deviation of 8%, a 16%- deficiency in cortical bone is significant, while with a standard deviation of 12% it 'is not. So this design principle gives the best sensitivity to disorders.

Notice that radiogrammetry measures the volume of bone tissue rather than its mineral content. If mineralization is a constant, as it appears to be in healthy subjects, this is the same thing. But some disorders alter the degree of mineralization and radiogrammetry is insensitive to this. This means that radiogrammetry is selectively sensitive to osteopenia (defined as decreased amount of bone tissue) and insensitive to osteomalacia [2] .

The relative standard deviations of the six candidate bone mass measures, averaged over age and gender bins, for the 629 children of the study was found to:

The combination A/ (LW) has the smallest relative standard deviation, and it is statistical significant that it is smaller than the other combinations. It is therefore the preferred measure of bone mass, and is called the Pediatric Bone Index (PBI) . The core of the invention is therefore the expression

PBI = A / (LW) , with A = π T W (1 - T/W)

The invention could also be useful in the adult population, in which case it is called Universal Bone Index.

The bone length could instead be represented by the arm length or body height, but it is preferred to derive it for the same bone and using the same equipment that measures the cortical thickness. This

has the advantage that PBI becomes insensitive to image magnification, which is always present in some degree in ordinary radiography.

The invention is also understood to include variations over these basic principles, which are obvious for a person skilled in the art, and which will have the same effect. For instance one can replace W measured in the shaft by the average width W of the bone from proximal to distal end; then LW is the projected area of the bone. Also, the bones used for the method which are preferably metacarpal 2-4 can be varied to other bones with a cortical shaft, e.g. the set of metacarpals in both hands. Also, any multiple of PBI would have the same effect as PBI, and the same is true for any monotonous function of PBI .

The invention does not, however, include the expression BMD/L, where BMD is determined by absorptiometry, for instance by using an aluminium wedge next to the hand during exposure with X-rays .

T and W are preferably computed as averages over a shaft region covering the middle 25% of the bone, where the middle is defined as the midpoint between the ends, but could be averaged over other parts of the shaft, e.g. defined in terms of the narrowest location on the bone .

Figure 3 shows PBI for boys and girls versus bone age (using plus signs) together with curves representing mean values (solid line) and 2 standard deviations (dotted lines) . The mean and std as a function of age or bone age is denoted "a reference database", or "normative data". By comparing a measurement on an individual with the reference data, one can make a statement of the degree of normality of the subject. For instance one can define +- 2 std as the normal range. Bone age is the apparent age of the child as judged from the morphology of the bones in a hand radiograph. Chronological age can be used instead of bone age as an approximation .

The PBI of non-dominant hands were compared to PBI of age- and gender-matched dominant hands. The non-dominant hands have 3% less PBI. The method can take this into account when comparing to the reference data.

Comparison with pQCT

PBI is remarkably constant before puberty, for girls even slightly decreasing. Then it increases at puberty. This development is similar to what is seen in pQCT of the distal radius recorded 5% from the end of radius. This measurement site has become popular in pQCT and normative data were published by Neu et al [3] . This is illustrated in figure 4, where the mean values of PBI in each age bin from the above mentioned clinical study are shown as points with error bars equal to the uncertainties of the estimate of these mean values. The normative pQCT data are shown as broken lines, and they have been scaled to fit the PBI data using the same constant for boys and girls. The agreement is good in the region 7-15 year, indicating that PBI is approximately equivalent to pQCT for this age group.

To replace the use of pQCT by PBI would be a great advantage due to the smaller radiation dose of PBI.

Computer implementation

The method can in a preferred embodiment be performed on a suitably programmed computer, or a system of several computers, which can be at separate physical locations and connected by a network, e.g. the Internet .

The method can thus be embodied as the computer program, the computer, the medium storing the computer program, or the interaction through a carrier wave mediating interaction over the Internet .

References

[1] DXR-BMD patent US 6,763,257 Bl, by Rosholm and Thodberg. (This patent contains a lot of references and general discussion which serves as background for the present application)

[2] E Schδnau et al: From Bone Biology to Bone Analysis, Hormone Research 2004, 61, p 257-269 (2004)

[3] CM Neu et al: Bone Density and Bone Size at the Distal Radius in Healthy Children and Adolescence, A study using Peripheral Quantitative Computed Tomography, Bone, 28, p 227-232 (2001) .

[4] H. H. Thodberg: Hands-on Experience with Active Appearance Models, Medical Imaging 2002: Image Proc, Eds. Sonka & Fitzpatrick, Proc. SPIE VoI. 4684, 495-506 (2002).