Regularization techniques in realistic Laplacian computation

IEEE Trans Biomed Eng. 2007 Nov;54(11):1993-9. doi: 10.1109/TBME.2007.893496.

Abstract

This paper explores regularization options for the ill-posed spline coefficient equations in the realistic Laplacian computation. We investigate the use of the Tikhonov regularization, truncated singular value decomposition, and the so-called lambda-correction with the regularization parameter chosen by the L-curve, generalized cross-validation, quasi-optimality, and the discrepancy principle criteria. The provided range of regularization techniques is much wider than in the previous works. The improvement of the realistic Laplacian is investigated by simulations on the three-shell spherical head model. The conclusion is that the best performance is provided by the combination of the Tikhonov regularization and the generalized cross-validation criterion-a combination that has never been suggested for this task before.

Publication types

  • Research Support, Non-U.S. Gov't

MeSH terms

  • Algorithms*
  • Brain / physiology*
  • Brain Mapping / methods*
  • Computer Simulation
  • Diagnosis, Computer-Assisted / methods*
  • Electroencephalography / methods*
  • Humans
  • Models, Neurological*