Learning and Nonlinear Models - Revista da Sociedade Brasileira de Redes Neurais, Vol. 1, No. 2, pp. 122-130, 2003
© Sociedade Brasileira de Redes Neurais

4. Conclusion

It was presented a different approach for tomographic reconstruction process involving neural network. Calculations to include
fan-beam geometry and interpolation were adapted for this approach. The proposed partially connected neural network doesn’t
need to be trained and its weights are previously calculated according to the geometry of the problem. The relationship
between the projection space and final image is purely geometric not only for parallel beam but also for fan-beam
configuration. Once image size and number of projections are determined the network can be assembled and any tomographic
image can be reconstructed from its projections presented to the network input.

Preliminary simulations on sequential processor showed 4 to 5 times speed-up in the reconstruction of 10 slices
compared to conventional backpropagation algorithm. This technique is sought to be very advantageous for neural parallel
hardware implementation where it will reach extremely high speed which is well desired for multi-slice reconstruction and real
time tomography.

References

[1] M.T. Munley et al. An artificial neural network approach to quantitative single photon emission computed tomographic
reconstruction with collimator, attenuation and scatter compensation. Med. Physics. 21 (12) (1994).

[2] J. P. Kerr and E. B. Bartlett. A statistical tailored neural network approach to tomographic image reconstruction. Med.
Physics. 22(5) (1995).

[3] R. G. S. Rodrigues. Desenvolvimento e Aplicaçao de um Algoritmo de Reconstruçao Tomogmfica com Base em Redes
Neurais Artificiais. Doctorate thesis. Faculdade de Filosofia, Ciências e Letras de Ribeirao Preto, SP (2000).

[4] H. Monma, Y. Chen, Z. Nakao. Computed Tomography base on a Self-organizing Neural Network, IEEE SMC '99
Conference Proceedings. (1999).

[5] L. F. Medeiros, H. P. da Silva and E. P. Ribeiro. Reconstruçao de Imagens Tomograficas Utilizando Redes Neurais
Parcialmente Conectadas. Proceedings of the V Brazilian Conference of Neural Networks. Rio de Janeiro, RJ (2001).

[6] L. F. Medeiros. Reconstruçao de Imagens Tomograficas com Redes Neurais Parcialmente Conectadas. Dissertaçao de
Mestrado. UFPR. Curitiba, PR (2001).

[7] A. Rosenfeld and A. Kak. Digital Picture Processing. Academic Press. San Diego (1982).

[8] L. F. Medeiros, H. P. da Silva and E. P. Ribeiro. Reconstruçao de Imagens Tomograficas Geradas por Projeçoes Fan-
Beam com Detectores Eqüidistantes usando Redes Neurais Parcialmente Conectadas. Proceedings of the VI Brazilian
Conference of Neural Networks. Sao Paulo, SP (2003).

[9] P. Williams, T. York. Hardware implementation of RAM-based neural networks for tomographic data processing. IEE
Proceedings of Computers and Digital Techniques, pp. 114-118, (1999).

[10] A. K. Jain. Fundamentals ofDigital Image Processing. Prentice-Hall, pp.431-475 (1989).

[11] G. T. Herman. Image Reconstruction from Projections. Academic Press, pp.90-160 (1980).

[12] G. T. Herman., S. W. Rowland and M. M. Yau. A comparative study of the use of linear interpolation and modified cubic
spline interpolation for image reconstruction. IEEE Transactions Nuclear Sciences, pp.2879-2894 (1973).

[13] S. Haykin. Neural Networks - A Comprehensive Foundation. MacmillanColl. Pub. Com. Inc. pp.1-41 (1994).

[14] A. Kak and M. Slaney. Principles of Computerized Imaging. IEEE Press. New York (1988).

130

<<   6   7   8   9  

More intriguing information