WO2008065394A1

WO2008065394A1 - Method and apparatus for reducing distortion in a computed tomography image

Info

Publication number: WO2008065394A1
Application number: PCT/GB2007/004557
Authority: WO
Inventors: Julian J. Liu
Original assignee: Isis Innovation Limited
Priority date: 2006-11-28
Filing date: 2007-11-28
Publication date: 2008-06-05
Also published as: GB0623720D0

Abstract

A method and apparatus for reducing artefact caused by a metal implant, thorax/pelvic streaking, and lower dose streaking in reconstruction images comprising the steps or means for pre-process the observed projection for baseline correction, performing an initial reconstruction of the object using an iterative process, generating ideal correspondence of the implant; locating the region affected by the implant, detecting and analysing inconsistencies caused by the implant inside a scanned object, approximating the real correspondence of the implant and updating said initial reconstruction, isolating the approximated correspondence of the implant, synthesising the implant-free reconstruction and the location of said implant, analysing reliability of the updated reconstruction, and displaying the reliability analysis as pixels of the reconstruction. The key techniques in this invention including an errorless forward modelling of CT scanning, beam hardening analysis and approximation, and noises identification and isolation prior to reconstruction.

Description

METHOD AND APPARATUS FOR REDUCING DISTORTION IN A COMPUTED

TOMOGRAPHY IMAGE

Background of the Invention

The present invention relates to a method and apparatus for reconstructing an image for X- ray computed tomography robust to metal artefact, thorax/pelvic streaking and lower dose streaking, with particular but not necessarily exclusive, emphasis on one or more of errorless forward modelling, beam hardening analysis and approximation, and noise identification and isolation prior to reconstruction.

Metal implants are one of the main reasons for the occurrence of image artefact, which is regarded as one of the key factors impeding post-operative evaluation using Computed Tomography (CT). Since the attenuation coefficient of metal is much higher than that of human tissue, a metal implant causes a low signal to noise ratio, beam hardening and scatter, all of which contribute to the projection data received from CT scans to be corrupted, by virtue of the production of a star-burst artefact and thus presenting an obstacle to the proper reconstruction of the image. The filtered back-projection (FBP) method is the most widely used method for image reconstruction since it gives high computational efficiency whilst maintaining reasonable accuracy [I]. Star-burst artefacts have been reduced by substituting the data corresponding to the projection lines through metal objects with data from the neighbourhood [2], or synthetic data using linear, polynomial, or other interpolation strategies [3,4], or by adjusting the wavelet decomposition coefficients [5]. These methods are very effective for removing streaking generated by metal implants and do not increase the amount of computation significantly. Iterative algorithms have been regarded as a potential method of providing high quality CT reconstructions, especially for metal artefact reduction [6-8]. An alternative strategy for avoiding metal artefact was investigated by using a low attenuation material such as titanium [9]. Currently, the standard reconstruction strategy in clinical practice is still a filtered back projection with linear interpolation, which is reasonable for some general applications.

However, in some situations such as planning a metal implant replacement operation and post-operation observation, it is extremely important to know the exact structure of the tissues surrounding the metal implant. Using filtered back projection (FBP) with linear interpolation or nearest neighbour interpolation, the reconstruction image in the area near the boundary of the metal implant is most likely to be distorted. It is true that the information is still sufficient if the affected scan data is ignored, but the FBP does not allow such a gap, which must be interpolated. These strategies inevitably cause some information loss and can still result in artefacts.

Conventional CT scanners suitable for clinical radiology do not possess a function allowing a metal artefact reduction when scanning patients with metal implants. When a metal implant is present in the sample, which is regarded as the key factor impeding a post-operation evaluation, the reconstruction is seriously distorted by artefacts and no method has yet been devised that successfully meet the clinical demands.

The above-mentioned problem has been a matter of discussion for over nearly three decades and the few successful techniques, such as the use of an iterative method, have involved unacceptable computational costs. The iterative method ignores the metal affected data successfully, so that high quality images can be obtained, but some areas may fail to be reconstructed due to ignorance. The computational expense is another problem in respect of which much effort has been focused on this subject. Using a low attenuation material is not ideal for clinical applications because X-ray attenuation characteristics should not be the primary factor in material selection.

Accordingly, the present inventor has realised that a new method for computed tomography, parallel and equal to Radon transformation but with pixel voxel based description based on sinusoidal rather than line integrals could achieve the reduction of metal artefacts without the burden of computational costs, as well as eliminate streaking in upper-thorax/pelvic imaging and lower dose imaging to improve the quality of image.

Summary of the Invention

It is an object of the present invention to solve the above problems, avoid the image reconstructions of patients with metal implant being distorted, and eliminate streaking in upper-thorax/pelvic imaging and lower dose imaging. A first specific object of the present invention is to reduce the artefact caused by the metal implant during CT scanning.

A second specific object of the present invention is to improve significantly the quality of the scanned image.

To attain the above objects, according to the first aspect of the present invention, there is provided a method for reducing distortion caused by an implant and/or thorax /pelvic streaking in a CT scanned image of an object, the method comprising the steps of : performing an initial reconstruction of the object using an iterative process; locating the region affected by an implant ; detecting and analysing inconsistencies caused by the implant inside said scanned object; approximating the real correspondence of the implant, updating said initial reconstruction; and isolating the approximated correspondence of the implant, updating said initial reconstruction and isolating the approximated correspondence of the implant; synthesising the implant- free reconstruction and the location of said implant; analysing reliability of the updated reconstruction; and displaying the reliability analysis as pixels of the reconstruction..

Further, according to the second aspect of the present invention, there is provided an apparatus for reducing distortion caused by an implant and/or thorax/pelvic streaking in a CT scanned image of an object, the apparatus being arranged and configured to comprise the steps of: perform an initial reconstruction of the object using ain iterative process; locate the region affected by the implant; detect and analyse inconsistencies caused by the implant inside the scanned object; approximate the real correspondence of the implant, update said initial reconstruction, and isolate the approximated correspondence of the implant; synthesise the implant-free reconstruction and the location of said implant; analyse reliability of the updated reconstruction; and display the reliability analysis as pixels of the reconstruction.

Brief Description of Drawings

FIG. 1 is an analytically geometrical description of errorless forward modelling.

FIG 2 shows a diagram of the method and apparatus of the present invention.

FIG 3 (a) shows an observed projection from a conventional CT scanner.

FIG 3 (b) shows a default reconstruction of a conventional CT scanner.

FIG 4 depicts the inconsistency of the projection of a conventional CT scanner.

FIG 5 (a) shows an ideal correspondence of the metal implant in the projection.

FIG 5 (b) shows the maximum value of each view of the ideal correspondence of the metal implant, which strongly correlates to the inconsistency description shown in FIG 4.

FIG 6 (a) depicts the approximated correspondence of the metal implant.

FIG 6 (b) depicts the compensation for beam hardening.

FIG 7 depicts the detected high energy noises in projection.

FIG 8 (a) is a projection amended using the method and apparatus of the present invention.

FIG 8 (b) is a reconstruction using the method and apparatus of the present invention. FIG 9 (a) is a sagittal reconstruction of the original output of a conventional CT scanner.

FIG 9 (b) is a sagittal reconstruction using the method and apparatus of the present invention, where clearer image has been obtained.

FIG 10 (a) is a default output of a conventional CT scanner for pelvic imaging.

FIG 10 (b) is a reconstructions using the method and apparatus of the present invention, where streaking has been eliminated.

FIG 11 (a) is a default output of a CT scanner for lower dose imaging.

FIG 11 (b) is a reconstruction of the same lower dose scan using the method and apparatus of the present invention, where the image quality has been significantly improved.

Below, preferred aspects of the present invention are explained with reference to the accompanying drawings.

The method according to the present invention comprises the steps of first pre-processing raw data from the scanner to obtain a projection and a multiple-order (2-7) polynomial is automatically identified to form a baseline and to correct the projection.

In each X-ray view, the attenuation values of a projection smoothed using a multiple phase wavelet decomposition are accumulated to detect an initial global inconsistency baseline. In order to identify and correct inconsistencies caused by X-rays travelling through an examination table used for carrying patients into the scanner (also called "couch"), data from this local couch-caused inconsistency, which is normally detected in a small area where the attenuation has a local maximum, is accumulated in the direction of detectors.

The inconsistency caused by the scanned body (also called "scanning-object-inconsistency description") is then obtained by accumulating a smoothed projection in the direction of detectors, where the initial global inconsistency baseline, the local couch-caused inconsistency, and high energy noise caused by metal implant, detected using multiple phase wavelet decomposition, are removed. During the initial reconstruction, the location of the metal implant is first identified and then the correspondence of the metal implant is generated so the region affected is detected. The initial reconstruction is obtained by iterating several times and analysing the errors.

The ideal correspondence of the metal implant in the projection data set is a group of extended sinusoidal curves and is generated by an errorless forward modelling algorithm to be used as a basic function to approximate beam hardening.

Let PQ denote the original projection collected from a CT scanner, R be the operation of a reconstruction algorithm, F_M be the operation of forward modelling, and C₁= R (Po).

The original projection can therefore be represented as

P₀ = F(CO + n (1)

where r_/ are the residuals, which are ideally expected as white noise. However, due to the limitation of the reconstruction algorithm and ill-posed problem, rj may include more information rather than just white noise.

The filter back projection, for example, aims at reducing the accumulated error causing the boundary of different tissues to blur, however, the effect could not be completely obtained. There may be some information in rj that should have been extracted by reconstruction, and can be retrieved by repeating the algorithm of reconstruction. An iterative algorithm can be effective to overcome this kind of limitation of reconstruction algorithm, which is described as follows:

Theoretically, equals to since F is a linear operator. In

practice, the computation of ill cause accumulating errors if using a normal discrete algorithm of forwar modelling.

Equivalently, equation (2.k) can be presented as

normal discrete forward modelling

(3) errorless forward modelling

where r_k is supposed not to contain information that can be extracted by reconstruction.

The iterative can be terminated when is smaller than a given positive small number

or start to oscillate after a dramatic decreasing period, and the output is expressed as

If the residuals P₀ - F(C) are white noise according to local and global statistical evaluation, then the reconstruction can be regarded as reliable, otherwise further analysis will be needed.

When metal implants exist, the residuals Po - F(C) are most likely not white noise due to beam hardening and high energy noise. In this situation, the reconstruction is an ill-posed inverse problem. The reconstruction will not be completely reliable because it may be based on the wrong information, and the objective of the reconstruction can be a minimum variance of the residual.

There are two key factors causing inconsistency: beam hardening and high energy noise caused by scattering. The region in projection affected by metal implants and its neighbourhood is with high probability of the presence of high energy noise, and the region in projection affected by metal implants and the views through and parallel or approximately parallel to the couch is probably affected by beam hardening. Accordingly, the projection can be divided into a high reliability region and a low reliability region.

Beam hardening is regarded as one of the key factors causing artefact and impeding a detailed post operation evaluation using CT, when metal implants are present. Ideally, the projection of CT scan is the integration alone the X-ray beam path, but because of high attenuation, the beam of X-ray becomes harder and causes nonlinearities, especially in the region affected by metal implants.

The accuracy of the forward modelling is sensitive to the approximation of beam hardening. An errorless forward modelling algorithm is implemented by extending the sinusoidal description, combining analytical computation with discrete computation, and substituting a pixel/voxel with a unit square/cube.

Errorless forward modelling for parallel beam is described as follows, which can be easily extended to fan beam, axial cone beam, and spiral cone beam. If the descriptions of the errorless forward modelling for fan beam, axial cone beam, and spiral cone beam are interested, please contact the author.

Let a unit square present a pixel in an axial structure, with the co-ordinates of the centre as (x,y) and (x_c,y_c) be the scan centre.

At the view of θ , the X-rays through this unit square can be detected by one, or two, or even three detectors, and the attenuation values can be calculated according to the corresponding areas of the unit square determined by the angle of the view θ^* and the distance between the mapping of the centre of the unit square and the boundary of the detector.

Since a unit square is an axial/central symmetrical geometry, it will be completely overlapped if rotating every π/2. Therefore a substitution of #^*can be derived as The centre of the unit square will be

mapped on detector corresponding to

The location where the centre of the unit square is mapped on the corresponding detector can be represented by the distance to the boundaries of the detector a and

is floor function of /. The unit square is therefore divided into four parts: the area above 1 +

denoted as Si, the area between / and 1

denoted as S₂i, the area betwee nd /, denoted as .S₂₂, and the area below denoted as S₃. The

attenuation values on the three corresponding detectors are S_\, S₂₂, and Si, respectively.

The values of Si, S₂₁, S₂₂, and S₃ can be calculated as follows, (see figure 1).

The output of this forward modelling combines analytical computation with discrete computation, without alias and errors caused by interpolation. Therefore the ideal correspondence of the initially detected metal implant is generated using this method of forward modelling accurately.

The ideal correspondence of the initially detected metal implant and inconsistency description of the noise free projection are then analysed according to the distribution of measurement functions.

Let P_ιcm be the ideal correspondence of the initially detected metal implant, and F_ιcm is the measurement function of P_ιcm , expressed as

where is a positive integer (e.g.

K_1n =IOOO ), N_det and N_vιews denote the number of the detectors and the number of the views, respectively.

Let P_smw be the smoothed projection using multiple phase wavelet decomposition, S_psmw(j) be the summation of the smoothed detected values in jth view, expressed as

and F_lp be the measurement function

^QfS_ps_mΛJ) .expr

Let V_ιcm {k) denote the views containing the areas of P_ιcm (i,j) ≥ k - max(P_lcm )/ K_m , and

V (k) be the views that

There will be a group of pairs of integer number, where

such that V_lcm (&,) is completely overlap

Wύh

If , then the beam hardening will not be so serious that no

area is completely fail to attenuate the X-ray. In such a situation, a kernel function can be identified with features as

where α and x₀ are parameters to be identified.

If there i a such that when and

whe then k_r is a threshold value that having travelled through

metal material with the thickness as k,. ^■ max(P_Icm ) / K_m , the X-ray will not be attenuated anymore due to beam hardening.

The approximation base function is

where - max represents the

compensation for beam hardening caused by metal implant, and Q denotes the compensation of beam hardening for the biological tissues in the shadow of metal implant. The parameter k_r ■ max can be directly computed, and the parameters of x₀ and

a are automatically identified with a large set of initial values to avoid local minimum. The number of the initial values is sensitive to computational efficiency, as well as the accuracy of the approximation. Therefore the number of the initial values needs to be balanced with computational cost. The supervision function plays a key role to the control of beam hardening approximation, which is formed by combining the summation of the correspondence of the scanning-object in the direction of detectors and the moving average values.

With the substitution of the identified parameters expressed in the above formula, the error caused by beam hardening can be adjusted, where x is the ideal correspondence of the metal implants, and K_β is the approximated compensation value corresponding to beam hardening. The projection is amended by the compensation, and reconstructed similar to the initial reconstruction discussed above. The detected locations of metal implant are compared with the previous detections which will be updated if different, and repeat the steps following initial reconstruction.

The region affected by the metal implants and its neighbourhood can be smoothed by decomposing the projection with the isolation of the approximated real correspondence of the metal implants into several components using wavelet in MATLAB toolbox. In this application, only the two highest frequency bands are focused on, and the phase of the wavelet base function in the highest frequency band and the second highest frequency band are shifted V₂ and ¹A, respectively. There are four components in the highest frequency band, and 16 components in the second highest frequency band.

In each component, the high reliable region and the low reliable region are analysed statistically, and a function can be derived to map the variance and the distribution in the low reliable region into the same levels as in the high reliable region. The statistically amended components corresponding to each phase of wavelet functions are reconstructed separately. An amended projection can be generated by the synthesis of all the wavelet reconstructions, and the high energy noises caused by the metal implant are eliminated.

For upper thorax and pelvic imaging, noises are most likely generated in projection when X-rays travel long distance through biological tissues. Phantom for upper thorax/pelvic can be scanned, as well as simulated using errorless forward modelling to generate references for noise analysis. The smoothed projection can be decomposed into subsets according to the measurement function. In each subset the noses are statistically analysed, identified, and shrunk in wavelet domain. The streaking can then be eliminated from reconstruction. This method can be directly extended to lower dose imaging.

The final, the reconstructed image is obtained by synthesizing the implant -free reconstruction of the amended projection data and the position of the metal implant. This procedure has the potential to be developed into a new reconstruction algorithm, based on errorless forward modelling. The reconstruction is processed to match the expected attenuation values and output with dicom format. As a last step, the reliability analysis is displayed as pixels of the reconstruction.

The method is described with a diagram in Figure 2.

Three examples those show the advantage of the present invention over former methods are described below.

The projections are collected from a CT scanner, GE Lightspeedlβ.

Example 1, a patient had both knees replaced was scanned with the parameters as 12OkV and 100mA.

The projection is presented in Figure 3(a), and the corresponding reconstruction from the CT scanner is shown in figure 3(b), where the boundary between the metal implants and human tissues is blurring. There is streaking in the reconstruction, and some areas adjacent to the metal implant are dark caused by beam hardening.

The inconsistency can be shown by summing up the attenuation values in each view in the smoothed projection, and part of the inconsistency is caused by beam hardening, as shown in figure 4.

The ideal correspondence of the metal implant in projection is generated by errorless forward modelling, based on the initial reconstruction using the global iterative, as shown in figure 5(a). The maximum of the ideal correspondence of the metal implant in each view is shown in figure 5(b), where strong similarity with the inconsistency description shown in figure 4 can be detected.

The approximated correspondence of the metal implant in projection and the beam hardening compensation are presented in figure 6 (a) and (b), respectively.

The high energy noises are detected using multiple phase wavelet decomposition, as shown in figure 7. The approximated correspondence of the metal implant and the detected high energy noises are isolated from the projection, as shown in figure 8 (a). The amended projection is reconstructed in figure 8(b), where the metal implant and the adjacent area are clearly imaged.

The whole series is processed using the method and apparatus of the present invention, and a saggital reconstruction is presented in figure 9(b), while the corresponding original is in figure 9(a). For comparison, the original output of the CT scanner failed to provide sufficient information, and obstructed the diagnosis.

Example 2, a pelvic imaging

A patient was scanned for pelvic imaging with parameters as 14OkV and 32OmA. The default output of the CT scanner is shown in Figure 10(a), where streaking in horizontal direction affects the image. Prior to reconstruction, noises in the projection are analysed, identified, and shrunk using the method and apparatus of the present invention. The streaking has been eliminated, and the quality of the image has been obviously improved, as shown in Figure 10(b).

Example 3, lower dose imaging

A CT performance phantom 76-410 was amounted in a plastic container filled with water and Niopam 300 iodinated contrast, 10 ml per litter, to simulate upper thorax or pelvic imaging. The scanning parameters are 14OkV and 80mA, much lower than a normal clinical application.

The default output of the CT scanner is shown in Figure 11 (a), where serious streaking takes place, especially in diagonal directions. The reconstruction could hardly reveal details of the geometry. The reconstruction using the method and apparatus of the present invention is shown in Figure l l(b), where the image quality has been significantly improved, and the geometry has been clearly imaged. REFERENCES

1. R. H. Bracewell, and A. C. Riddle, "Inversion of fan beam scans in radio astronomy," Astrophysics J., Vol. 150, pp.427 - 434,1967.

2. G. H. Glover, and N. J. PeIc, "An algorithm for the reduction of metal clip artefacts in CT reconstructions," Med. Phys., Vol.8, pp799-807, 1981.

3. W. A. Kalender, R. Hebel, and J. Ebersberger, "Reduction of CT artefacts caused by metallic implants," Radiology, Vol. 164, pp576-577, 1987.

4. A. H. Mahnken, R. Raupach, J. E. Wildberger, B. Jung, N. Heussen, T. Flohr, R. W. Gϋnther, and S. Schaller, "A new algorithm for metal artefact reduction in computed tomography," Investigative Radiology, Vol. 38, pp769-775, 2003.

5. S. Zhao, D. D. Robertson, G. Wang, and B. Whiting, "X-ray CT metal artefact reduction using wavelets: an application for imaging total hip prostheses," IEEE Trans. Med. Imag, Vol. 19, ppl238 - 1247, 2000.

6. J. Rockmore, and A. Macovsky, "A maximum likelihood approach to emission image reconstruction from projections," IEEE Trans. Nucl. Sci., Vol. 23, ppl428- 1432, 1976.

7. C T. Byrne, "Convergent block- iterative algorithm for image reconstruction from inconsistent data," IEEE Trans. Imag. Proc, Vol. 6, ppl296-1304, 1997.

8. J. Liu, S. R. Watt-Smith, and S. M. Smith, "A Description of Computed Tomography based on Sinusoidal Curves," J. X-ray Sci. Tech., Vol. 11, pp205-218, 2003.

9. J. Liu, S. A. Billings, Z. Zhu, and J. Shen, Enhanced Frequency Analysis using Wavelets, International Journal of Control, Vol. 75(15), PP.l 145-1158, 2002.

Claims

CLAIMS:

1. A method for reducing distortion caused by an implant and/or thorax /pelvic streaking in a CT scanned image of an object, the method comprising the steps of : performing an initial reconstruction of the object using an iterative process; locating the region affected by an implant ; detecting and analysing inconsistencies caused by the implant inside said scanned object; approximating the real correspondence of the implant, updating said initial reconstruction; and isolating the approximated correspondence of the implant, updating said initial reconstruction and isolating the approximated correspondence of the implant; synthesising the implant- free reconstruction and the location of said implant; analysing reliability of the updated reconstruction; and displaying the reliability analysis as pixels of the reconstruction.

2. A method according to claim 1, including the further step of pre-processing raw scanned data to obtain a projection and forming an initial correction baseline, prior to said step of performing said initial reconstruction.

3. A method according to claim 1 or claim 2, including the step of identifying and eliminating noise.

4. A method according to any one of claims 1 to 3, including the step of generating an ideal correspondence of said implant.

5. A method according to claim 2, wherein a multiple-order polynomial is automatically identified to form the initial correction baseline.

6. The method according to any one of claims 1 to 5, wherein the implant is a metal implant.

7. A method according to claim 1, wherein the location of the implant is detected according to said initial reconstruction.

8. A method according to claim 4 wherein the step of generating the ideal correspondence of the metal implant is performed by means of errorless forward modelling after the step of locating the region affected by the implant.

9. A method according to claim 8, where the errorless forward modelling method is developed by combining analytical and numerical computation for simulating two or more of parallel beam, fan beam, cone beam axial, and cone beam spiral scanning.

10. A method according to any one of the preceding claims, further comprising the steps of computing a global inconsistency description of a filtered projection using a multiple phase wavelet decomposition.

11. A method according to any one of the preceding claims, further comprising the step of analysis of inconsistency based on the distribution of the measurement function of the ideal correspondence of the implant and the said global inconsistency description.

12. A method according to any one of the preceding claims, further comprising the step of determination of the structure of the model for inconsistency approximation.

13. A method according to any one of the preceding claims, further comprising the step of identification of parameters for the model with determined structure.

14. A method according to any one of the preceding claims, further comprising the step of comparing the reconstruction of the amended projection with the said initial reconstruction, update the said initial reconstructions and may repeat the corresponding steps depending on the comparison.

15. A method for clinical radiology according to any one of the preceding claims, further comprising the steps of smoothing reference projection generation; observed projection decomposition and statistical analysis; shrinking components corresponding to noises; synthetic reconstruction from wavelet domain.

16. A method according to claim 15, wherein a reference projection is generated by scanning a phantom as well as forward modelling the specification of the phantom using the said errorless forward modelling.

17. A method according to claim 15, wherein a database of reference noise distribution of projection can be built up for clinical applications.

18. A method according to claims 15 wherein said reference noise distribution of the projection is generated from areas of projection with lower energies of noise according to the analysis of the distribution of measurement function of the projection if neither phantom nor reference database is available.

19. A method according to any one of the preceding claims, wherein an algorithm is used to shrink the noise in wavelet domain according to the noise distribution in wavelet domain and corresponding references.

20. The method according to any one of the preceding claims, wherein the projection is updated by a synthetic reconstruction from shrunk components in wavelet domain.

21. An apparatus for reducing distortion caused by an implant and/or thorax/pelvic streaking in a CT scanned image of an object, the apparatus being arranged and configured to comprise the steps of: perform an initial reconstruction of the object using an iterative process; locate the region affected by the implant; detect and analyse inconsistencies caused by the implant inside the scanned object; approximate the real correspondence of the implant, update said initial reconstruction, and isolate the approximated correspondence of the implant; synthesise the implant-free reconstruction and the location of said implant; analyse reliability of the updated reconstruction; and display the reliability analysis as pixels of the reconstruction.

22. Apparatus according to claim 21 , further arranged and configured to pre-process raw scanned data to obtain a projection and forming an initial correction baseline, prior to performance of said initial reconstruction.

23. Apparatus according to claim 21 or claim 22, further arranged and configured to identify and eliminate noise.

24. Apparatus according to any one of the clams 21 to 23, further arranged and configured to generate an ideal correspondence of said implant.

25. Apparatus according to claim 22, wherein a multiple-order polynomial is automatically identified to form the initial correction baseline.

26. Apparatus according to any one of claims 21 to 25, wherein the implant is a metal implant.

27. Apparatus according to any one of claims 21 to 26, wherein the location of the implant is detected according to the said initial reconstruction.

28. Apparatus according to any one of claims 21 to 27, further arranged and configured to generate the ideal correspondence of the metal implant by errorless forward modelling after locating the region affected by the implant.

29. Apparatus according to claim 28, wherein errorless forward modelling method is developed by combining analytical and numerical computation for simulating two or more of parallel beam, fan beam, cone beam axial and cone beam spiral scanning.

30. Apparatus according to any one of claims 21 to 29, arranged and configured to compute a global inconsistency description of a filtered projection using a multiple phases wavelet decomposition.

31. Apparatus according to any one of claims 21 to 30, further arranged and configured to analyse inconsistency based on the distribution of the measurement function of the ideal correspondence of the implant and said global inconsistency description.

32. Apparatus according to any one of claims 21 to 31 , further arranged and configured to determine the structure of the model for inconsistency approximation.

33. Apparatus according to claim 32, further arranged and configured to identify parameters for the model with determined structure.

34. Apparatus according to any one of claims 21 to 33 further arranged and configured to compare the reconstruction of the amended projection with the said initial reconstruction, update the said initial reconstruction, and repeat as required depending on said comparison.

35. Apparatus for clinical radiology according to any one of claims 21 to 34, arranged and configured to smooth reference projection generation; observe projection decomposition and statistical analysis; shrink components corresponding to noises; and perform synthetic reconstruction from wavelet domain.

36. Apparatus according to claim 35, wherein a reference projection is generated by scanning a phantom as well as forward modelling the specification of the phantom using the said errorless forward modelling.

37. Apparatus according to claim 35, wherein a database of reference noise distribution of projection can be built up for clinical applications.

38. Apparatus according to claim 35, wherein a reference noise distribution of the projection is generated from areas of projection with lower energies of noise according to the analysis of the distribution of measurement function of the projection if neither phantom nor reference database is available.

39. Apparatus according to any one of claims 21 to 38, wherein an algorithm is used to shrink the noise in wavelet domain according to the noise distribution in wavelet domain and corresponding references.

40. Apparatus according to any one of claims 21 to 39, wherein the projection is updated by a synthetic reconstruction from shrunk components in wavelet domain.