Skip to main content
Log in

Non-Iterative Hierarchical Registration for Medical Images

Journal of Signal Processing Systems Aims and scope Submit manuscript

Abstract

Digital images from diverse medical imaging modalities and from different imaging times are becoming an indispensable information resource for making clinical decisions. Image registration is an enabling technique for more fully utilizing the embedded heterogeneous image information. However, in addition to the complex differences and deformations inherent in the medical images, the increasing scope, resolution, and dimensionality of imaging pose significant challenges in this medical arena. Wavelets have shown great potential in multi-scale registration due to their superior capacity for representing image information at different resolutions and spatial frequencies. However, the application of wavelets in registration is hindered by their lack of rotation- and translation-invariance. To overcome this obstacle, this paper proposes a non-iterative hierarchical registration method based on points of interest which are extracted automatically from wavelet decompositions. The proposed algorithm for two-dimensional monomodal medical images has been validated by experiments on phantom data and clinical imaging data. This proposed non-iterative method provides a computationally efficient registration, as well as assists in avoiding the non-convergence problem.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Institutional subscriptions

Figure 1
Figure 2
Algorithm 1
Figure 3
Figure 4
Figure 5
Figure 6
Figure 7
Figure 8
Figure 9
Figure 10
Figure 11
Figure 12

References

  1. Wang, X., Stefan, S., Fulham, M., Som, S., & Feng, D. (2007). Data registration and fusion, Chapter 8. In D. Feng (Ed.), Biomedical information technology. Amsterdam: Elsevier.

    Google Scholar 

  2. Brown, L. G. (1992). A survey of image registration techniques. ACM Computing Surveys, 24(4), 325–376.

    Article  Google Scholar 

  3. Maintz, J. B. A., & Viergever, M. A. (1998). A survey of medical image registration. Medical Image Analysis, 2(1), 1–36.

    Article  Google Scholar 

  4. Fitzpatrick, J. M., Hill, D. L. G., & Maurer, C. R. (2000). Handbook of medical imaging, vol.2, Ch. Image Registration pp. 375–435. Bellingham: SPIE.

    Google Scholar 

  5. Mäkelä, T., Clarysse, P., Sipilä, O., Pauna, N., Pham, Q. C., Katila, T., & Magnin, I. E. (2002). A review of cardiac image registration methods. IEEE Transaction on Medical Imaging, 21(9), 1011–1021.

    Article  Google Scholar 

  6. Bankman, I. (2000). Handbook of medical imaging: processing and analysis. New York: Academic.

    Google Scholar 

  7. Pluim, J. P. W., Maintz, J. B. A., & Viergever, M. A. (2001). Mutual information matching in multiresolution contexts. Image and Vision Computing, 19(1–2), 45–52.

    Article  Google Scholar 

  8. Kovacic, S., & Bajcsy, R. (1998). Multiscale/multiresolution representation. In W. Toga (Ed.), Brain warping (pp. 45–65). New York: Academic.

    Google Scholar 

  9. Lester, H., & Arridge, S. R. (1999). A survey of hierarchical non-linear medical image registration. Patter Recognition, 32, 129–149.

    Article  Google Scholar 

  10. Thévenaz, P., & Unser, M. (2000). Optimization of mutual information for multiresolution registration. IEEE Transaction on Image Processing, 9(12), 2083–2099.

    Article  MATH  Google Scholar 

  11. Unser, M., & Aldroubi, A. (1993). A multiresolution image registration procedure using spline pyramids. In: A. F. Laine (Ed.), Proc. of SPIE vol. 2034, pp.160–170, Wavelet Applications in Signal and Image Processing, Nov 1993

  12. Brigger, P., Muller, F., Illgner, K., & Unser, M. (1999). Centered pyramids. IEEE Transactions on Image Processing, 8(9), 1254–1264.

    Article  Google Scholar 

  13. You, J., & Bhattacharya, P. (2000). A wavelet-based coarse-to-fine image matching scheme in a parallel virtual machine environment. IEEE Transactions on Image Processing, 9(9), 1547–1559.

    Article  Google Scholar 

  14. Gefen, S., Tretiak, O., Bertrand, L., Rosen, G. D., & Nissanov, J. (2004). Surface alignment of an elastic body using a multiresolution wavelet representation. IEEE Transactions on Biomedical Engineering, 51(7), 1230–1241 July.

    Article  Google Scholar 

  15. Wang, X., & Feng, D. (2005). An efficient wavelet-based biomedical registration for abdominal images. Journal of Nuclear Medicine, 46(Suppl), 161.

    Google Scholar 

  16. Daubechies, I. (1988). Orthogonal bases of compactly supported wavelets. Communications on Pure and Applied Mathematics, 41, 909–996.

    Article  MATH  MathSciNet  Google Scholar 

  17. Mallat, S. (1989). A theory of multiresolution signal decomposition: The wavelet representation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 11, 674–693.

    Article  MATH  Google Scholar 

  18. Mallat, S. (1989). Multiresolution approximations and wavelet orthonormal bases of Ĺ(R). Transactions of the American Mathematical Society, 315(1), 69–87.

    Article  MATH  MathSciNet  Google Scholar 

  19. Unser, M., Aldroubi, A., & Laine, A. (2003). Guest editorial: Wavelets in medical imaging. IEEE Transactions on Medical Imaging, 22(3), 285–288.

    Article  Google Scholar 

  20. Mallat, S., & Hwang, W. L. (1992). Singularity detection and processing with wavelets. IEEE Transactions on Information Theory, 38, 617–643.

    Article  MathSciNet  Google Scholar 

  21. Mallat, S. (1998). A wavelet tour of signal processing. San Diego: Academic.

    MATH  Google Scholar 

  22. Simoncelli, E. P., Freeman, W. T., Adelson, E. H., & Heeger, D. J. (1992). Shiftable multiscale transforms. IEEE Transactions on Information Theory, 38, 587–607.

    Article  MathSciNet  Google Scholar 

  23. Liang, J., & Parks, T. W. (1996). A translation-invariant wavelet representation algorithm with applications. IEEE Transactions on Signal Processing, 44, 225–232.

    Article  Google Scholar 

  24. Cole-Rhodes, A. A., Johnson, K. L., LeMoigne, J., & Zavorin, I. (2003). Multiresolution registration of remote sensing imagery by optimization of mutual information using a stochastic gradient. IEEE Transactions on Image Processing, 12(12), 1495–1511.

    Article  MathSciNet  Google Scholar 

  25. Tsai, D., & Chiang, C. (2002). Rotation-invariant pattern matching using wavelet decomposition. Pattern Recognition Letters, 23, 191–201.

    Article  MATH  Google Scholar 

  26. Xue, Z., Shen, D., & Davatzikos, C. (2004). Determining correspondence in 3-D MR brain images using attribute vectors as morphological signatures of voxels. IEEE Transactions on Medical Imaging, 23(10), 1276–1291.

    Article  Google Scholar 

  27. Fitzpatrick, J. M., West, J. B., & Maurer Jr., C. R. (1998). Predicting error in rigid-body, point-based registration. IEEE Transactions on Medical Imaging, 17(5), 694–702.

    Article  Google Scholar 

  28. Umeyama, S. (1991). Least-squares of transformation parameters between two point patterns. IEEE Transactions on Pattern Analysis and Machine Intelligence, 13(4), 376–380.

    Article  Google Scholar 

  29. Hsieh, J. W., Liao, H. Y. M., Fan, K. C., Ko, M. T., & Hung, Y. P. (1997). Image registration using a new edge-based approach. Computer Vision and Image Understanding, 67(2), 112–130.

    Article  Google Scholar 

  30. Wang, X., & Feng, D. (2005). Active contour based efficient registration for biomedical brain images. Journal of Cerebral Blood Flow & Metabolism, 25(Suppl), S623.

    Article  Google Scholar 

  31. Mallat, S., & Zhong, S. (1992). Characterization of signal from multiscale edges. IEEE Transactions on Pattern Analysis and Machine Intelligence, 14(7), 710–732.

    Article  Google Scholar 

  32. Bookstein, F. L. (1989). Principal warps: Thin-plate splines and the decomposition of deformations. IEEE Transactions on Pattern Analysis and Machine Intelligence, 11(6), 567–585.

    Article  MATH  Google Scholar 

  33. Rohr, K. (2001). Landmark-based image analysis: Using geometric and intensity models. Computational imaging and vision series, vol. 21. Dordrecht: Kluwer Academic.

    Google Scholar 

Download references

Acknowledgement

This work was supported by ARC and PolyU/UGC grants. The authors would like to express their appreciation to Dr. Ian Parkin for polishing the writing.

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Xiuying Wang.

Appendices

Appendix 1: Wavelet Transform

The continuous wavelet transform of a function \(f\left( x \right) \in L^2 \left( \Re \right)\) can be defined as [20]:

$$\begin{array}{*{20}l}{\left( {W_\Psi f\left( x \right)} \right)\left( {a,b} \right) = \left\langle {f\left( x \right),\Psi _{ab} \left( x \right)} \right\rangle } \hfill \\{ = \frac{1}{{\sqrt {\left| a \right|} }}\int_\Re {f(x)\Psi \left( {\frac{{x - b}}{a}} \right)} dx} \hfill \\\end{array} $$
(1)

where the function ψ ab (x) defines the family of the wavelet functions with \(a,b \in \Re \); and \(a \ne 0\) is the dilation parameter; and b is the translation parameter. The definition also provides the similarity measurement of f(x) and wavelets ψab(x) obtained from mother wavelet ψ(x) by dilation and translation:

$$\Psi _{ab} \left( x \right) = \sqrt {2^{ - a} } \Psi \left( {2^{ - a} x - b} \right)$$

Similarly, the discrete wavelet of a one-dimensional discrete function f(n) is defined as:

$$\left( {W_\Psi f\left( n \right)} \right)\left( {j,k} \right) = \sum\limits_i {f\left( n \right)\Psi _{jk} \left( n \right)} $$
(2)

where \(\Psi _{jk} \left( n \right) = \sqrt {2^{ - j} } \Psi \left( {2^{ - j} n - k} \right)\); n = 2j + k for \(j = 0,1,...,\log _2 \left( N \right) - 1\), \(k = 0,1,...,2^j - 1\).

The mother wavelet ψ(x) can be constructed from a scaling function ϕ(x): \(\varphi \left( x \right) = \sqrt 2 \sum\limits_n {h_\varphi \left( n \right)} \varphi \left( {2x - n} \right)\) and \(\Psi \left( x \right) = \sqrt 2 \sum\limits_n {h_\Psi \left( n \right)} \varphi \left( {2x - n} \right)\) where \(h_\varphi \left( n \right)\) is the impulse response of a discrete filter, which needs to meet certain requirements to ensure the set of basis wavelet functions to be orthonormal and unique; while \(h_\psi \left( n \right)\) can be extracted from \(h_\varphi \left( n \right)\): \(h_\Psi \left( n \right) = \left( { - 1} \right)^n h_\varphi \left( {L - 1 - n} \right)\), where L is the length of \(h_\varphi \left( n \right) \cdot h_\varphi \left( n \right)\) and \(h_\psi \left( n \right)\) correspond to lowpass and highpass filters.

For fast discrete wavelet transform, the series of filtering and down-sampling operations used to compute \(W_\psi \left( {j,m,n} \right)\) and \(\left\{ {W_\Psi ^i \left( {\,j,m,n} \right)|i = D,V,H} \right\}\) can be mathematically expressed as:

$$\begin{array}{*{20}c} {W_{\Psi } {\left( {j,m,n} \right)} = h_{\varphi } {\left( m \right)}*{\left[ {h_{\varphi } {\left( n \right)}*W_{\Psi } {\left( {j - 1,m,n} \right)}|_{{n = 2k,k \geqslant 0}} } \right]}|_{{m = 2k,k \geqslant 0}} } \\ {W_{\Psi } ^{H} {\left( {j,m,n} \right)} = h_{\Psi } {\left( m \right)}*{\left[ {h_{\varphi } {\left( n \right)}*W_{\Psi } {\left( {j - 1,m,n} \right)}|_{{n = 2k,k \geqslant 0}} } \right]}|_{{m = 2k,k \geqslant 0}} } \\ {W_{\Psi } ^{V} {\left( {j,m,n} \right)} = h_{\varphi } {\left( m \right)}*{\left[ {h_{\Psi } {\left( n \right)}*W_{\Psi } {\left( {j - 1,m,n} \right)}|_{{n = 2k,k \geqslant 0}} } \right]}|_{{m = 2k,k \geqslant 0}} } \\ {W_{\Psi } ^{D} {\left( {j,m,n} \right)} = h_{\Psi } {\left( m \right)}*{\left[ {h_{\Psi } {\left( n \right)}*W_{\Psi } {\left( {j - 1,m,n} \right)}|_{{n = 2k,k \geqslant 0}} } \right]}|_{{m = 2k,k \geqslant 0}} } \\ \end{array} $$
(3)

where \(W_\psi \left( {j,m,n} \right)\) is approximation coefficients used to represent global (low frequency) information; \(W_\Psi ^H \left( {\,j,m,n} \right)\), \(W_\Psi ^D \left( {\,j,m,n} \right)\), and \(W_\Psi ^V \left( {\,j,m,n} \right)\) are horizontal coefficients, diagonal coefficients, and vertical coefficients respectively, which represent local (high frequency) information.

Appendix 2: Thin-plate Spline

The thin-plate spline interpolation was firstly introduced by Bookstein [32] into medical image registration area and since then, it has been one of commonly used elastic registration methods. Rohr gave a good study about elastic registration based on TPS [33].

The POI of reference image and study image are respectively \(P = \left\{ {\left. {p_i = \left( {x_i ,y_i } \right)} \right|i = 1,2,......n} \right\} \in I_R \) and \(Q = \left\{ {\left. {q_i = (x_i^\prime ,y_i^\prime )} \right|i = 1,2,......n} \right\} \in I_S \). To map the corresponding POI and to produce a smooth interpolation, a transformation function \(f:f\left( {\,p_i } \right) = q_i ,i = 1,2,...,n\) is to be determined to minimize the energy function E which reflects the amount of variation. The regulation parameter λ > 0 is used to control the fitness. The energy function can be expressed as:

$$\begin{array}{*{20}l}{E\left( f \right) = \sum\limits_{i = 1}^n {\left\| {q_i - \left. {f\left( {p_i } \right)} \right\|} \right.} ^2 } \hfill \\{ + \lambda \iint\limits_\infty {\left( {\left( {\frac{{\partial ^2 f}}{{\partial x^2 }}} \right)^2 + 2\left( {\frac{{\partial ^2 f}}{{\partial x\partial y}}} \right)^2 + \left( {\frac{{\partial ^2 f}}{{\partial y^2 }}} \right)^2 } \right))dxdy}} \hfill \\\end{array} $$
(4)

The thin-plate spline interpolation function can be written as:

$$f\left( {x,y} \right) = \sum\limits_{k = 1}^3 {a_k } \phi _k \left( t \right) + \sum\limits_{i = 1}^n {w_i U\left( {\left\| {\left( {x,y} \right) - p_i } \right\|} \right)} $$
(5)

where the coefficients a 1, a 2 and a 3 define the affine part of the transformation, and \(\left[ {\phi _1 \left( t \right),\phi _2 \left( t \right),\phi _3 \left( t \right)} \right] = \left[ {1,x,y} \right]\); whereas the coefficient w defines the elastic deformation, and \(U\left( {\left\| {\left( {x,y} \right) - p_i } \right\|} \right) = \left\| {\left( {x,y} \right) - p_i } \right\|^2 \log \left( {\left\| {\left( {x,y} \right) - p_i } \right\|} \right)\).

In order to keep f(x,y) having square integrable second derivatives, the following conditions must be satisfied: \(\sum\limits_{i = 1}^n {w_i = 0} \) and \(\sum\limits_{i = 1}^n {w_i x_i = } \sum\limits_{i = 1}^n {w_i y_i = 0} \).

The coefficient vector a = (a 1,a 2,a 3)T and w = (w 1,w 2,...w n )T can be computed through the following linear equations:

$$\left\{ {\begin{array}{*{20}c}{kw + Pa = v} \\{P^T w = 0} \\\end{array} } \right.$$
(6)

where v represents column vectors of landmarks; \(k_{ij} = U_i \left( {p_j } \right) = U\left( {\left\| {\left( {x_i ,y_i } \right) - \left( {x_j ,y_j } \right)} \right\|} \right)\); and (1,x,y) is the ith row in P. These two vector equations can be solved by:

$$\left\{ {\begin{array}{*{20}c}{w = K^{ - 1} (v - Pw){\text{ }}} \\{a = (P^T K^{ - 1} P)^{ - 1} P^T K^{ - 1} v} \\\end{array} } \right.$$
(7)

Rights and permissions

Reprints and permissions

About this article

Cite this article

Wang, X., Feng, D.D. Non-Iterative Hierarchical Registration for Medical Images. J Sign Process Syst Sign Image Video Technol 54, 65–77 (2009). https://doi.org/10.1007/s11265-008-0183-9

Download citation

  • Received:

  • Revised:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s11265-008-0183-9

Keywords

Navigation