Journal of Radioanalytical and Nuclear Chemistry, Vol. 269, No.2 (2006) 325–329
0236–5731/USD 20.00 Akadémiai Kiadó, Budapest
© 2006 Akadémiai Kiadó, Budapest Springer, Dordrecht
Curve fitting using a genetic algorithm for the X-ray fluorescence
measurement of lead in bone
L. Luo,
1,2*
D. R. Chettle,
2
H. Nie,
2
F. E. McNeill,
2
M. Popovic
2
1
National Research Center of Geoanalysis, 100037, Beijing, P.R. China
2
Department of Medical Physics and Applied Radiation Sciences, McMaster University, Hamilton, Canada
(Received April 6, 2006)
We investigated the potential application of the genetic algorithm in the analysis of X-ray fluorescence spectra from measurement of lead in bone.
Candidate solutions are first designed based on the field knowledge and the whole operation, evaluation, selection, crossover and mutation, is then
repeated until a given convergence criterion is met. An average-parameters based genetic algorithm is suggested to improve the fitting precision
and accuracy. Relative standard deviation (RSD%) of fitting amplitude, peak position and width is 1.3–7.1, 0.009–0.14 and 1.4–3.3, separately.
The genetic algorithm was shown to make a good resolution and fitting of K lines of Pb and γ elastic peaks.
Introduction
In the in vivo X-ray fluorescence measurement of
trace lead in bone, continuous efforts are made to lower
the minimum detectable limit of lead thus improving the
measurement precision particularly in individuals with
low exposure.
1
Such improvements are partially limited
by the energy resolution of the XRF system, which can
be insufficient to resolve the overlap of lead K-series X-
rays and the Compton peak. The resolution of
overlapped peaks may provide a robust foundation of
predicting accurately the concentrations of trace Pb in
phantoms and in bone.
A trend in research involving curve fitting is to find a
global optimum solution and to avoid falling into a local
minimum. The genetic algorithm (GA) was first
suggested by HOLLAND in 1975,
2
and since then genetic
algorithms have been extensively studied and have
found a wide range of applications.
3–7
It has been
confirmed that GAs are a useful tool in solving real
problems in many scientific fields, involving
optimization and curve fitting. GAs are also of great
value in identifying the fittest model by natural
evolution. Generally, GAs have good search accuracy,
but are flawed by poor precision. Thus, improving the
search precision is a task that requires addressing before
GAs can be reliably applied to X-ray spectrometry.
In this work, a GA has been investigated and used to
resolve the overlapping peaks in the X-ray fluorescence
measurement of lead in bone. Special attention has been
paid to improving fitting precision.
Theory
GA is based on the principle of biological evolution
and survival of the fittest. Entities capable of adapting to
the environmental changes are able to live longer. GA
* E-mail: luoliqiang@ccsd.org.cn
pays attention to the evolution of biological genus in
order to avoid the degradation of a system.
GA usually includes the initialization of data,
evaluation of fitting and genetic operations. A
convergence criterion is assigned prior to the evaluation.
Candidate solutions, or chromosomes, are first designed
based on the field knowledge and then are initialized
randomly. The initial solutions or parameters are limited
within upper and low boundaries. Then they are
forwarded to genetic operations to go through selection
and crossover procedures. The optimal solutions are
kept and worse ones excluded. The whole operation,
evaluation, selection, crossover and mutation, is then
repeated until a given convergence criterion is met.
In principle, the crossover and mutation are the
mathematical operations of exchanging positions or
values of parameters fitted.
In general, criteria are the number of generations, or
small error values predefined by a user. The population
size is usually from 50 to 500 strings, the crossover
probability is in the region 0.5–0.9 and the mutation
probability is between 0.001 and 0.05.
The key difference between GA and common
classical methods are that GA approaches search whole
space of solutions rather than a small range of
parameters as done by the common methods.
Experimental
Two major cases of overlapped peaks are
encountered when Pb in bone is measured by XRF. One
is due to the coherent scatter peak from
109
Cd source,
which overlaps nearly completely with Pb Kβ2.
Secondly, there is an overlap between Pb peaks and the
Compton peak, which causes the main obstacle in the
determination of trace Pb in bone. Pb Kα
1
peak at trace
content, a prominent lead feature, is difficult to
distinguish from a large tail of Compton peak.