在膝關節表面重建手術中，由於所裝配的植入物表面直接與病患的半月板及關節軟骨接觸摩擦，若此一表面與膝關節未受損處之表面無法正確吻合，則可能導致病患膝關節的損傷及加速植入物的損壞。為了避免上述問題，本研究希望藉由膝關節上受損部位周圍之軟骨表面資料，重建受損部位應有的平滑表面，以做為植入物設計的參考。
在表面重建的方法上，本研究提出以移動式最小平方法和徑向基底函數神經網路作為修補膝關節表面缺損的方法。前者是在常見的最小平方法中加入權重值做調整，以修正嵌合的誤差；後者則是利用所建構之徑向基底函數，以函數逼近的方式找到輸入與輸出間的關係，藉此建立膝關節表面的數學模型。
由實驗結果可以發現徑向基底函數神經網路較適合應用於修補膝關節表面上的缺損，在膝關節較易受損的內髁部位，即使受損部份佔重建區域達49.7％，其修補的最大誤差僅為0.171㎜、平均誤差亦僅為0.052㎜。故日後在設計用於修補膝關節表面微小受損的植入物時，可考慮採用以徑向基底函數神經網路計算所得之曲面作為植入物設計的依據。

In resurfacing reconstruction of the knee joint, the surface of the implant will contact the plateau and joint cartilage dynamically. If the implant surface does not fit well with the joint surface, it could lead to the damage of both the knee joint and the implant.
This research aims to use the surrounding cartilage surface of the joint defect to reconstruct original surface, which will be used for the design of the resurfacing implant.
As for the methods of surface reconstruction, moving least squares method and radial basis function neural network are proposed for restoration of the normal surface of the joint defect. The former uses a weighting method to compensate the errors of fitting; while the latter uses radial basis functions to find the relationship of input and output in order to modeling the surface. Both of the methods are applied to reconstruct various curved surfaces. The experimental results show that radial basis function neural network is better to restore the damaged knee joint surface. Also, at the easily damaged condyle part of the knee joint, the maximum approximation error is 0.171㎜ and the average approximation error is only 0.052㎜ even if the defect part of the reconstruction area reaches 49.7%. Based on the results, the radial basis function neural network is recommended to be used to generate resurfacing implants for the reconstruction of the knee joint defect.

[1] Hoppe, H., DeRose, T., Duchamp, T., et al., “Surface reconstruction from unorganized points,” SIGGRAPH’92, pp. 71-78, 1992.
[2] Lorensen, W. E., Cline, H. E., “Marching cubes:a high resolution 3D surface construction algorithm,” Computer Graphics, Vol. 21, pp. 163-169, 1987.
[3] Dobkin, D.P., Levy, S.V.F., Thurston, W.P., et al., “Contour tracing by piecewise linear approximations,” ACM TOG, Vol. 9, pp. 389-423, 1990.
[4] Gopi, M., Krishnan, S., “A fast and efficient projection based approach for surface reconstruction,” Journal of High Performance Computer Graphics, Multimedia and Visualisation, Vol. 1, pp. 1-12, 2000.
[5] Amenta, N., Bern, M., Kamvysselis, M., “A new Voronoi-based surface reconstruction algorithm,” SIGGRAPH’98, pp. 415-421, 1998.
[6] Jun, Y., “A piecewise hole filling algorithm in reverse engineering,” Computer-Aided Design, Vol. 37, pp. 263-270, 2005.
[7] Davies, J., Marschner, S.R., Garr, M., et al., “Filling holes in complex surface using volumetric diffusion,” First International Symposium on 3D Data Processing, Visualization, Transmission, 2002.
[8] Wang, J., Oliveira, M.M., “A hole-filling strategy for reconstruction of smooth surface in range image,” XVI Brazilian Symposium on Computer Graphics and Image Processing, Geometric Modeling 1, pp. 11-18, 2003.
[9] Lancaster, P., Salkauskas, K., “Curve and surface fitting, an introduction,” Academic Press, 1986.
[10] Rumelhart, D.E., Hinton, G.E., Williams, R.J., “Learning internal representation by error propagation,” Parallel Distributed Processing, Vol. 1, pp. 318-362, 1986.
[11] Lazzaro, D., Montefusco, L.B., “Radial basis function for the multivariate interpolation of large scattered data set,” Journal of Computational and Applied Mathematics, Vol. 140, pp. 521-536, 2002.
[12] Carr, J.C., Fright, W.R., “Surface interpolation with radial basis function for medical imaging,” IEEE Transaction on Medical Imaging, Vol. 16, pp. 96-107, 1997.
[13] Levin, D., “The approximation power of moving least-squares,” Mathematic of Computation, Vol. 67, pp. 1517-1531, 1998.
[14] Powell, M.J.D., “Radial basis function for multivariable interpolation: a review,” IMA Conference on Algorithms for Approximation of Functions and Data, pp. 143-167, 1987.
[15] Chen, S., Cowan, C.F.N., Grant, P.M., “Orthogonal least squares learning algorithm for radial basis function networks,” IEEE Transactions on Neural Networks, Vol. 2, pp. 302-309, 1991.
[16] Franke, R., “Smooth interpolation of scattered data by local thin plate splines,” Computers and Mathematics with Applications, Vol. 8, pp. 273-281, 1982.
[17] Powell, M.J.D, “The theory of radial basis function approximation in 1990,” Advances in Numerical Analysis II: Wavelets, Subdivision Algorithms, and Radial Basis Functions, Oxford University Press, pp. 105-210, 1990.
[18] Niyogi, P., Girosi, F., “On the relationship between generalization error, hypothesis complexity, and sample complexity for radial basis functions,” Neural Computation, Vol. 8, pp. 819-842, 1996.
[19] Gupta, M.M., Jin, L., Homma, N., “Static and dynamic neural network,” John Wiley & Sons, Inc, pp. 223-246, 2003.
[20] Xia, Q., Wang, M.Y., Wu, X., “Orthogonal least squares in partition of unity surface reconstruction with radial basis function,” International Conference on Geometric Modeling and Imaging (GMAI 2006), pp. 28-33, 2006.