1Philips Medical Systems Division, Building QA-II, P.O. Box 218, NL-5600 MD Eindhoven, The Netherlands
2University of Nijmegen, The Netherlands
Abstract
Based on the spline software of Lyche et al.,¹ a subroutine package is presented in which the amount of smoothing on a set of n noisy datapoints is determined from the data by means of the Generalized Cross-Validation (GCV) or predicted Mean-Squared Error (MSE) criteria of Wahba² and her collaborators. Following an idea of Hutchinson and de Hoog,³ an efficient O (m²n) algorithm is used for calculating the criterion functions, where 2m is the order of the spline function. In this fashion, earlier O (n³) approaches based on the singular value decomposition can be avoided.