EP1771819A1

EP1771819A1 - Method and device for improving perceptibility different structures on radiographs

Info

Publication number: EP1771819A1
Application number: EP04741280A
Authority: EP
Inventors: Michael Thoms
Original assignee: Duerr Dental SE
Current assignee: Duerr Dental SE
Priority date: 2004-07-27
Filing date: 2004-07-27
Publication date: 2007-04-11
Also published as: US8244019B2; JP4755184B2; EA011562B1; WO2006010372A1; KR20070049177A; US20090169086A1; KR101036370B1; JP2008507998A; EA200700347A1; CN101052992A; CN101052992B

Abstract

The invention relates to a method for improving the perceptibility of different structures (18) on radiographs (10) by means of an image processing device (20) consisting a) in storing a radiograph (10) in electronic form as a local space-intensity distribution, b) in carrying out a Fourier transformation for determining a frequency-intensity distribution, c) in filtering said frequency-intensity distribution by modifying weighting between the high-frequency and low-frequency image signal components, wherein the fixing of the image signal components to be more intensively weighted is carried out taking into account the mean structure size of said structures whose perceptibility is to be improved, d) in carrying out an inverse Fourier transformation of the filtered frequency-intensity distribution in order to obtain a modified space-intensity distribution in which said structures are more easily perceptible. The image contrast for the hardly perceptible structures may be selectively improved by means of a changed weighting for the high-frequency, relative to the low-frequency, image signal components in the Fourier spectrum because the structures hardly perceptible on the radiographs, for example soft tissue parts having a size quantity and structuring different from easily perceptible structures as bones and implantants.

Description

Method and device for improving the visibility of different structures on radiographic images

The invention relates to a method for improving the recognizability of different structures on radiation images as well as a suitable image processing device.

In medical diagnostics, the evaluation of radiographic images such as X-ray images is of great importance. Bones, implants or similar structures are generally distinct from surrounding soft tissues and are therefore easily recognizable. Conversely, soft tissue structures, such as tendons or blood vessels, are usually only very vaguely reflected on radiographic images. In numerous clinical pictures, however, the recognition of the soft-tissue structures is or is particularly important. In addition, it is often difficult to distinguish similar tissue from each other. For example, smaller bones, which are imaged on a ^radiographic image over a larger bone, are often barely visible to the naked eye; the same applies to partial soft tissue structures. In such cases, therefore, physicians can often make no or only a rather unreliable diagnosis on the basis of the radiation images alone. A certain improvement has been created by the digitization of radiographic images. By known methods of image processing, for example the contrast enhancement within selected image sections, soft tissue structures, for example, may under certain circumstances be emphasized more clearly. As a rule, however, this does not make it possible to detect a tendon lying above a bone. Namely, the smaller variations in the signal level of image signal components representing the tendon can not significantly emerge before the high background signal level of the bone. Although an image screen used for the display shows the slight fluctuations in the signal level even in the best case as intensity fluctuations, these are usually so small that they are hardly recognizable to the naked eye.

The object of the invention is therefore to indicate a method and a device for improving the recognizability of structures of different types on radiographic images.

This task is solved by a procedure with the

Features of claim 1 and by a Vorrich¬ device with the features of claim 9.

The invention is based on the finding that in most cases the structures whose recognizability is to be improved, in terms of their size and fineness of the other on the radiographic image mapped Differentiate structures more or less clearly. Since smaller and finer structures in the Fourier spectrum are represented by higher frequencies than large, coarse structures, a change in the weighting between high-frequency and low-frequency Bildsignalantei¬ len in the Fourier spectrum amplification of the image contrast can be achieved either for small fine or large coarse structures. Depending on whether the structures, which are difficult to recognize, are finer or coarser than the structures that are easily recognizable, the weighting of the image signal components in the

Frequency space changed either in favor of high-frequency or low-frequency image signal components.

The structures, which are at first difficult to recognize, emerge particularly clearly when the image signal components to be weighted are defined according to claim 2.

In the particularly simple filtering according to claim 3, the frequency space intensity distribution is simply multiplied by a filter function.

By using frequency center values and profile functions for determining the frequency ranges to be weighted according to patent claims 4 and 5, the filtering can be specifically influenced with relatively few parameters.

According to patent claim 6, a Gaussian function is particularly suitable as a profile function since it has the property of a Gaussian function even in the case of the Fourier inverse transformation to stay. The filtering can then be represented in the spatial domain as a convolution of the intensity distribution with a Gaussian function. This prevents the filter from flowing apart in the image as a result of which the intensity distribution abruptly changes and thus has a particularly high contrast.

Which frequencies or frequency ranges are changed in their weighting is determined by the average structure size of the structures whose recognizability is to be improved. The average structure size or corresponding frequency ranges can either be fixed or, according to claim 7, freely selectable by means of actuators on the image processing device or via a user interface of a superordinate computer. By changing the relevant filter parameters, a physician can thus improve the recognizability of the structures of interest to him in a variety of fluoroscopic images.

In addition, an automatic determination of the frequency ranges by means of an adaptive method according to claim 8 comes into consideration.

The selection of the structures, whose visibility is to be improved, can be done, for example, as indicated in the claims 9 or 10. An additional high-frequency filtering according to claim 11, for example with a Gaussian filter according to claim 12, leads to an increase in the signal-to-noise ratio, since image structures reproducing image signal components are amplified compared to a high-frequency background noise. Such filtering takes account of the fact that in the images frequently to be displayed in practice, the Fourier amplitudes decrease with increasing frequency f.

The advantageous embodiments and advantages cited above for the method also apply mutatis mutandis to the image processing device according to the invention.

Further features and advantages of the invention will become apparent from the following description of an exemplary embodiment with reference to the drawing. Show:

Figure 1 is an X-ray image on which a finger bone and soft tissue structures are visible;

Figure 2 is a block diagram of a Bild¬ processing apparatus according to the invention;

FIG. 3 shows a one-dimensional periodic spatial intensity distribution I (x);

FIG. 4 shows the frequency-space intensity distribution F (f) of the spatial-space intensity distribution I (x) from FIG. 3; FIG. 5 shows a filter as shown in FIG

Frequency-space intensity distribution F (f) generates filtered frequency-space intensity distribution F '(f);

FIG. 6 shows a spatial-space intensity distribution I '(x) obtained by Fourier inverse transformation of the filtered frequency-space intensity distribution F ¹ (f) shown in FIG. 5;

FIG. 7 shows a frequency-space intensity distribution F (f) of a one-dimensional spatial-space intensity distribution with two profile functions gi (f) and g ₂ (f);

FIG. 8 shows a filter as shown in FIG

Frequency-spread intensity distribution F (f) obtained filtered frequency-space intensity distribution F '(f).

FIG. 1 shows a typical X-ray image 10 of a finger 12, on which several finger bones 14 as well as the surrounding soft tissue 16 can be seen. Because of their high density in relation to the soft tissue 16, the finger bones 14 are distinguished from it with high contrast, while soft tissue structures such as tendons 18 can hardly be seen on the X-ray image 10. A diagnosis of ^¬-like structures of the soft tissue 16 using the Rönt¬ genbildes 10 is therefore tet behaf- with larger uncertainty. _ η

Even if the conventionally recorded X-ray image 10 is digitized in a scanner and displayed on a screen with high contrast, the soft tissue structures 18 remain difficult to recognize. The reason for this is that image signal components reproducing the finger bones 14 or other high-density hard tissue have a very high signal level. Smaller fluctuations in the signal level, which represent the soft tissue structures 18 of interest, are hardly significant in relation to the high signal levels of the finger bones 14. Although a high-quality screen may still reproduce the slight fluctuations in the signal level as intensity fluctuations, these are then so small that they are barely visible to the naked eye. The same applies, moreover, to PSL image plates (PSL = photostimulable luminescence) which are not chemically developed, but in which the X-ray image contained therein must be read on a screen by opto-mechanical scanning prior to viewing.

In order to improve the recognizability of the soft tissue structures 18, the digitized X-ray image 10 is therefore processed in an image processing device 20, the structure of which is shown in FIG. The image processing apparatus 20 comprises a memory MEM in which the digital image data generated in a scanner SCAN can be stored. The digital image data represent intensities I (x, y) for each pixel P = (x, y), whereby, for example, each pixel P can be coded with 16 bits, so that more than 65,000 brightness values can be distinguished.

The memory MEM is connected to a Fourier transformation unit FT, with which digital image data read from the memory MEM can undergo a Fourier transformation. The frequency-space intensity distribution F (f _x , f _y ) generated by the Fourier transformation unit FT is a complex function over the frequency space spanned by the coordinates f _x and f _y and illustratively has the meaning of an amplitude density spectrum.

The image processing device 20 also comprises a filter unit FIL, with which the frequency space intensity distribution F (f _x , f _y ) can be filtered in such a way that the weighting of different frequency ranges is changed. This will be explained in more detail below with reference to Figures 3 to 6.

Finally, the image processing apparatus 20 has a Fourier-back transformation unit FT ^"1 , which transforms the frequency space intensity distribution F '(f _{x /} f _y ) filtered by the filter unit FIL back into the spatial domain, as a result of which a modified spatial intensity intensity distribution I An output 22 of the image processing device 20 can be used to connect a screen 24 on which the modified spatial intensity distribution I '(x, y) can be displayed. The filtering of the frequency space intensity distribution F (f _x , f _y ) in the filter unit is explained in more detail below with reference to FIGS. 2 to 6.

FIG. 3 shows an intensity distribution I (x) in the spatial domain for an image coordinate x, with a periodic distribution being assumed for the sake of simplicity. The intensity distribution I (x) represents a superimposition of a large-scale cosinusoidal intensity distribution with the period Pi with a small-scale cosinusoidal intensity distribution with the period P _2. The large-scale intensity distribution in this simplified example is assumed to be cosinusoidal here The shape of the bone is reproduced, while the small-scale intensity distribution represents the cosinusoidal form of cosinusoidal tissue structures arranged above it in the direction of transillumination, whose characteristic dimensions are significantly smaller than those of the bones. A half period length, ie a wave peak of the cosine function, corresponds in each case to the characteristic structure size.

As can be seen in FIG. 3, in front of the relatively high signal level of the large-area intensity distribution assigned to the bone, the small-scale intensity distribution associated with the soft tissue structures so goes into the background that the relatively small fluctuations of the small-scale intensity distribution become barely recognizable. if the overall intensity distribution I (x, y) shown in FIG. 3 would be displayed on a screen.

FIG. 4 shows the frequency-space intensity distribution F (f) for the spatial-space intensity distribution I (x) shown in FIG. The frequency-space intensity distribution F (f) has a contribution to the amount fi and a contribution to the amount f ₂ in addition to a contribution reproducing a DC component of the intensity distribution I (x) at the frequency f = 0, the smaller frequency fi the large-scale portion of the spatial intensity distribution and the larger frequency f _{2 represents} the small-scale portion of the spatial intensity distribution.

The filtering of the frequency-space intensity distribution F (x) is now performed so that the amplitudes of the contributions with the frequency amount fi reduced and the amplitudes of the contributions with the frequency amount f _{2 are} increased. This can be achieved, for example, by the following operations:

F '(fj = n ^• F (Jr ₁ ) and (1)

F '(f ₂ ) = r ₂ ^• F (f ₂ )

where .T ₁ and r _{2 are} gain factors with ri> 1 and r ₂ <1. The filtered frequency-space intensity distribution F '(f) obtained by the filtering is shown in FIG. The weighting of the image signal components with the frequency f _x and the image signal components with the frequency f ₂ is, as a comparison with FIG. 4 shows, in favor of the hochfrequenteren image signal component of the frequency f _{2 has been} changed verän¬.

FIG. 6 shows the modified intensity distribution I ¹ Ix) which is obtained by Fourier inverse transformation from the filtered frequency-space intensity distribution F '(f). It can be clearly seen in the illustration that the small-scale fluctuations in the intensity which the soft tissue structures are to reproduce now have a greater amplitude than before the filtering and therefore stand out much better from the large-scale fluctuations that represent the bones.

The example described above, by restricting it to cosinusoidal structures in only one dimension, represents a very crude simplification, from which, however, the nature of the filtering emerges particularly vividly. In real images, however, the structures shown have an arbitrary course over a wide range, which is why the frequency-space intensity distribution obtained by Fourier transformation represents a continuous function in frequency. If only the amplitudes of individual frequencies were then raised or lowered, as is the case in the above-described example, this only had an insignificant effect on the resulting filtered image.

For this reason, the weighting of the image signal ^components preferably takes place not only for individual discrete frequencies. zen, but for frequency bands. Each frequency band which is to be changed in its weighting is determined by means of a profile function. A Gaussian function is particularly suitable as a profile function since this has the property of retaining the shape of a Gaussian function even after the Fourier inverse transformation. A weighting of the image signal components by multiplication of the frequency space intensity distribution by a Gaussian function thus corresponds with a Gaussian function in the spatial domain of a convolution of the intensity distribution I (x, y). This in turn leads to the consequence that locations where the intensity distribution changes abruptly and which thus have a particularly high contrast do not appear to spatially disperse after filtering.

FIG. 7 shows the frequency-space intensity distribution F (f) for an arbitrary one-dimensional spatial-space intensity distribution, ie, not composed of cosinusoidal distributions. Plotted dashed lines are a first and a second profile function gi (f) and g ₂ (f), which in both cases by the equation

g ₃ (f) = exp (- (^ - f) ² / 2w ₃ ² ) (2)

given Gaussian functions with the central value f _Z i or fzi and the distance Wi or w ₂ between the central value and the inflection point. The distances Wi and w ₂ are a measure of the width of the bell-shaped profile functions gi (f) and g ₂ (f). The filter acts in this example such that the Fourier amplitudes of frequencies which lie within the lying around the central value f _Zi profile curve gi (f) can be lowered abge. Fourier amplitudes of frequencies which lie within the profile curve g _Σ Cf lying around the central value f _Z 2 are, on the other hand, increased.

In detail, the filtering of the frequency space intensity distribution F (f) is carried out according to the equation

F '(f) - F (f) • τ _F (f), (3)

where T _F (f) is a filter function given by

T _F (f) = (1 + _ri - _gi (f)) (l + r ₂ -g ₂ (f)). (4)

The amplification coefficients ri and rΣ indicate how much the Fourier amplitudes are to be changed within the frequency ranges predetermined by the profile functions. In the example shown, ri> 0, since the Fourier amplitudes are supposed to be raised by the smaller frequency f _Zi . For the amplification coefficient r ≤ r, r ₂ <0, which leads to a reduction of the Fourier amplitudes.

The effects of the filtering shown in FIG. 7 on the frequency-space intensity distribution F (f) are shown in FIG. Fourier amplitudes for frequencies lying within the filter profiles gi (f) and gz (f) are indicated by the

Filtering with the filter function T _F (f) is lowered or lifted. In this way, large-scale structures of the finger bones 14 on the image shown in the screen 24 recede relative to small-scale structures such as the tendons 18, so that the latter can be better recognized by a physician.

It is apparent from equations (2) to (4) that the filtering of the frequency space intensity distributions F (f) is determined by the value pairs (f _Z j, Wj) with the aid of the profile functions in the illustrated example. In this case, as in the example explained above with reference to FIGS. 3 to 6, the central _values f _Zj are preferably to be selected such that the half period _lengths corresponding to these frequencies f _Zj are approximately of the order of magnitude of those structures determined by the Filtering should be highlighted or attenuated during image display. If these typical _{feature sizes are the} same for all conceivable applications, then the center _values f _zj and also the profile _widths W _{j can be} fixed in the image processing _device 10. However, these parameters are preferably freely selectable by the physician with the aid of operating elements 26, 28 provided on the image processing device 10 in order to improve the recognizability of the soft tissue structures. Depending on the magnification, kind and arrangement of the bones of a hand, and the nature of the to be examined soft tissue structures other hand, may each cha ^¬ acteristic dimensions, may be quite different, so that an adjustment while watching the screen 24 will usually lead to the best results , The image signal components to be changed in their weighting can alternatively also be determined independently by the image processing device 20 by means of an adaptive method. For this purpose, it is to be determined by a doctor to be treated, which soft tissue structures are to be displayed better recognizable. For this purpose, the doctor preferably selects an edge region of the relevant soft tissue structure and marks it. FIG. 1 shows by way of example a marking designated by 30 which the physician can generate, for example, with a cursor on a screen used to display the X-ray image 10 and, in the illustrated example, comprises two points and a line 32 connecting them.

The image processing device 20 then executes the above-explained filtering for a multiplicity of frequency ranges and checks in each case to what extent the contrast along the line 32 between the points of the marking 30 is thereby improved. The modified intensity distribution obtained from that filtering, in which the highest contrast was achieved, is then displayed.

In the examples described so far been addressed thereof ausge¬ that by filtering in only one frequency range Fre¬ the Fourier amplitudes enlarged and in Ie diglich a frequency range, the Fourier amplitudes ver ^¬ be downsized. However, to improve the recognizability, only the amplitude ratio is important, so that in principle, one of the above-mentioned measures for improving the recognizability is sufficient. On the other hand, it may also be expedient to change Fourier amplitudes in more than two frequency ranges in order to achieve the desired improvement of the representation. For the counter j in Equations (2), (3) and (4), this means that it can assume not only the values 1 and 2 but also larger values.

The example explained above in connection with FIGS. 7 and 8 is limited to a one-dimensional intensity distribution I (x) for the purpose of simplifying the illustration. For image processing, the equations given above must be extended to two dimensions. Equation (3) can then be written taking into account equations (2) and (4) as

The filter function T _F (fχ, f _y ) can also be multiplied by another profile function to improve the signal-to-noise ratio. If it is w in die¬ ser profile function such as a Gaussian function with the Zen¬ tralfrequenz f _z = 0 and a width which corresponds to the frequency at which the high-frequency component frame signal buried in noise, so the image signal components against ^¬ are amplified above a background noise , This choice of profile function takes account of the fact that in the images which are frequently to be shown in practice, the Fourier Amplitudes decrease with increasing frequency f, so that at high frequencies usually always present noise signal outweighs.

It is understood that the above statements and explanations of an embodiment are limited only by way of example and in particular not to the improvement of the recognizability of soft tissue structures. As explained above, in the manner described above, not only soft tissue structures of hard tissue structures such as bone, but also similar Geweb¬ structures can be better distinguished from each other, provided that they differ in size from each other.

Claims

claims

1. A method for improving the recognizability of different structures (18) on radiographic images (10) using an image processing device (20), comprising the following steps:

a) storing a transmission image (10) in electronic form as a spatial distribution intensity distribution;

b) performing a Fourier transformation to determine a frequency domain intensity distribution;

c) filtering the frequency space intensity distribution by changing the weight between high-frequency and low-frequency image signal components, wherein the determination of the image signal components to be weighted more heavily takes into account an average structure size of the structures whose recognizability is to be improved;

d) carrying out a back-Fourier transformation of the filtered frequency-space intensity distribution, whereby a modified spatial-space intensity distribution is obtained, in which these structures are better recognizable.

2. The method of claim 1, wherein the determination of the more weighted image signal components such er¬ follows that the periods corresponding to these periods periods are approximately twice as large as the average structure size of the structures takes place, the visibility is to be improved.

3. The method of claim 1 or 2, wherein the filtered frequency space intensity distribution F '(f _x , f _y ) is given by

F '(4, ^) = T ^ f _x , f ^) • FIf _x , ^)

where F (f _x , f _y ) is the frequency space intensity distribution of the spatial intensity distribution, f _x , f _y frequencies in the two-dimensional frequency space and T _F (f _x , f _γ ) a Filterfunkti- on to weighting of image signal components.

4. The method according to any one of the preceding claims, wherein the determination of the filter function by at least one frequency center value and at least one profile function takes place which changes the weighting of the image signal components as a function of the distance to the frequency central value.

5. The method according to claims 3 and 4, wherein the filter function T _F (f _x , f _y ) is given by

where g (f _x , f _y , f _{Z j} , W ₃ ) is the profile function, r _{3 is} a weighting factor for the frequency center value f _2j and W _{j is} a width parameter which is a measure of the width of the profile function.

6. The method of claim 5, wherein the profile function g (f _{x /} f _y , fzj, Wj) by a

g (f _x , / 2w /)

given Gaussian function is.

7. The method according to any one of claims 4 to 6, wherein the determination of the at least one frequency Zentral¬ value and the at least one profile function with the aid of the image processing device (20) provided actuators (26, 28) is variable.

8. Method according to one of claims 4 to 6, in which the at least one frequency center value and the at least one profile function are so adaptively determined on the fluoroscopy image after a selection by a user of a structure whose identifiability is to be improved is that the contrast of this structure is increased.

9. The method according to claim 8, wherein the selection of the

Structure is done by specifying a point on the boundary of the structure and a direction along which the contrast is to be increased.

10. The method of claim 8, wherein the selection of the structure is performed by specifying two points within the structure, between which the contrast is to be increased.

11. The method according to any one of the preceding claims, wherein the frequency space intensity distribution F (f _{x /} f _y ) is additionally subjected to a high-frequency filtering.

12. The method of claim 11, wherein the Hochfrequenz¬ filter is given by a Gaussian filter with the frequency center value 0.

13. Image processing apparatus for improving the

Recognition of different structures (18) on radiographic images (10), with:

a) a memory (MEM) for storing a radiation image (10) present in electronic form as a spatial-space intensity distribution;

b) a Fourier transformation unit (FT) for carrying out a Fourier transformation for determining a frequency domain intensity distribution; c) a filter (FIL) for filtering the frequency space intensity distribution by changing the weighting between high-frequency and low-frequency Bildsi¬ gnalanteilen, wherein the determination of the more to be weighed image signal components taking into account a mean structure size of the structures whose visibility is to be improved;

d) a Fourier-back transformation unit (FT ^"1 ) for carrying out a back-Fourier transformation of the filtered frequency-space intensity distribution, whereby a modified spatial-space intensity distribution is obtained, in which these structures are better recognizable.

14. An image processing apparatus according to claim 13, wherein the _determining of the shares more to be emphasized Bildsignal¬ is performed such that the these frequencies entspre¬ sponding period lengths are at least approximately twice as large as the average feature size of the structures whose visibility should be improved.