TUTOR IAL

Limits of Detection in Spectroscopy

Volker Thomsen, Debbie Schatzlein, and David Mercuro

A diagrammatic view of the concept of detection limits is cL = 0  ksbiS, or

presented that effectively shows its relationship to the cL = ksbiS [1]

reproducibility of measurements on the blank.
This definition is admirably
illustrated in Figure 1 an ex-

A
pplications scientists A few comments are necessary ample from x-ray spectrometry
are constantly on the regarding this definition: which shows the results of
lookout for alterna- In the past, IUPAC has recom- some 30 separate, 20-s-long
tive ways to explain mended that k
3, although measurements on a 99.9% pure
some of the basic, but often con- many atomic absorption spec- iron sample. Measurements
fusing, concepts of spectrochem- troscopy and inductively cou- were made using a NITON
istry. Much has already been pled plasma optical emission model XLt 898 handheld x-ray
written about the subject of lim- spectroscopy practitioners use k fluorescence spectrometer with
its of detection, so this article
2 (2). miniature x-ray tube excitation
Volker will be short on words. Instead, IUPAC further suggests (NITON, Billerica, MA). The
Thomsen is we present diagrams that quite that, because the values for the silver anode tube was operated
senior effectively illustrate the concept blank, and the standard devia- at 35 kV and 5 A, with detec-
applications of limit of detection (LOD) and tion thereof, are estimates based tion by means of a silicon
scientist,
its relationship to precision. on limited measurements, the 3 positive-intrinsic-negative
Debbie
Schatzlein is sigma (3) value corresponds to (PIN) detector. We are measur-
senior Basic Definition a confidence level of about 90% ing the count rate of the copper
applications The International Union of Pure (3). K x-ray fluorescence with a
chemist, and and Applied Chemistry (IUPAC) The conversion from meas- peak region of interest (ROI) set
David defines LOD as follows (1): urement units, xL, to concentra- from about 7.85 to 8.25 keV.
Mercuro is
XRF marketing tion, cL, is via the sensitivity, S
The mean of the blank meas-
manager at The limit of detection, ex- concentration/intensity, the ures is xbi = 2.594 counts per
NITON LLC, 900 pressed as the concentra- slope of the calibration curve. second (cps), and the standard
Middlesex tion, cL, or the quantity, qL, That is, deviation of the 30 measure-
Turnpike, Bldg.
is derived from the smallest ments is sbi = 0.366 cps. Let k =
8, Billerica, MA
01821. Thomsen measure, xL, that can be de- cL
cbi  ksbiS 3 and we find xL = 2.594 
can be reached tected with reasonable cer- 3(0.366) = 3.692 cps. This limit
by e-mail at tainty for a given analytical But because the concentration of detection in intensity units is
vthomsen@ procedure. The value of xL of the blank is zero, this becomes indicated by the dotted line.
niton.com;
is given by the equation
Schatzlein can
be reached at xL
xbi  ksbi
Table I. Limits of detection and limits of quantitation.
dschatzlein@ where xbi is the mean of the
niton.com, and blank measures, sbi is the Absolute standard Relative standard
Mercuro can be standard deviation of the deviation deviation
reached at Limit of
blank measures, and k is a
dmercuro@
numerical factor chosen ac- detection 3 33%
niton.com.
cording to the confidence Limit of
level desired. quantitation 10 10%

112 Spectroscopy 18(12) December 2003 w w w. s p e c t r o s c o p y o n l i n e . c o m


5
Pure Fe: blank
Average blank
Limit of detection
4
Count rate (cps)
3
Count rate (cps)
2 3
1
0 0
0 4 8 12 16 20 24 28
Measurement number
Figure 1. Replicate intensity measurements on blank. Figure 2. Measurements on standard along with blank measurements.
Figure 2 shows the data of Figure 1 This example brings up several as-
along with 14 measurements on a low- pects of the limit-of-detection concept.
alloyed steel (C  12Mo) with 0.202%
copper. Limits of Detection and the BEC
Refer back to Figure 2. Note that the net
peak height is just about the same as the
A common misconception level of the background. But this is just
the definition of the background equiv-
is that the LOD is the alent concentration (BEC) (46). This
provides for an alternative method of
smallest concentration that computing the LOD, using the BEC and
the relative standard deviation (RSD) of
can be measured. Instead, the blank.
it is the concentration at cL
3 RSDblank BEC/100
which we can decide From the data on the blank provided
above (xbi
2.594 cps and sbi
0.366
whether an element is cps), we compute the RSD as 14.1%.
Therefore, cL
3(14.1)(0.2%)/100

present or not (6) that is, 0.085%. The slight discrepancy is due to
the point where we can our visual estimation of the BEC. If cal-
culated (5), the BEC is found to be
just distinguish a signal 0.19% and then the two computed lim-
its of detection become identical.
from the background.
Limits of Detection and Limits
of Quantitation
To find the sensitivity, we note that A common misconception is that the
the average of the 14 measurements is LOD is the smallest concentration that
5.325 cps copper. Thus our sensitivity is can be measured. Instead, it is the con-
S
(0.202%  0%)/(5.325 cps  2.594 centration at which we can decide
cps)
0.074%/cps. Multiplying 3 of whether an element is present or not
our blank and S, we arrive at a limit of (6) that is, the point where we can
detection for copper in an iron matrix just distinguish a signal from the back-
of 0.08%. ground. Quantitation is generally
Tutorial
7
6
5
4
2
1
0
4 8 12 16 20 24 28
Measurement number
agreed to begin at a concentration equal
to 10 standard deviations of the blank.
This is called the limit of quantitation
(LOQ) or limit of determination.
Therefore, LOQ
3.3 LOD. Thus, in
the example given in Figures 1 and 2,
the limit of quantitation is actually
about 0.24%.
Instrument Detection Limits Versus
Method Detection Limits
The example above provides the detec-
tion limit for copper in iron under
close-to-ideal conditions, where there
[2] are few other alloying elements. This is
an example of an instrument detection
limit (IDL) that is, a detection limit
in a clean matrix. The method detec-
tion limit (MDL) considers real-life
matrices. For example, we can expect
the LOD to degrade in stainless steels
containing several percent of nickel that
will raise the background in the copper
region of interest.
A rule of thumb typically used by op-
tical emission spectroscopists is that the
MDL can be anywhere from about two
to five times worse than the IDL. In
many x-ray fluorescence applications,
the IDLs are also typically a factor of
two to three times better than the
MDLs found with real-world samples.
However, this rule of thumb may not
necessarily apply across all spectroscopy
techniques and applications.
December 2003 18(12) Spectroscopy 113
Tutorial

35
Precision and the LOD 33% RSD
One final topic: Every practicing spec- 30
trochemist knows that as the concentra-

Relative standard deviation

tion of the analyte decreases, the preci-
sion, as expressed in the relative
standard deviation, gets worse. Quanti-
tatively speaking, the RSD is 10% at the
limit of quantitation and 33% at the
limit of detection (k = 3 in IUPAC
definition or 3 detection limits) (see
Table I).
Every practicing
spectrochemist knows that
as the concentration of the
analyte decreases, the
precision, as expressed in
the relative standard
deviation, gets worse.
Figure 3 shows this relationship be-
tween concentration and precision
graphically. Sixteen nickel-base alloys
with varying chromium content were
measured 11 times each for 20 s to
gather the precision data. Measure-
ments were made using the handheld x-
ray fluorescence spectrometer and
25
20
y
4.2086x0.6765
15
R2
0.9594
10
5
0
0 5 10 15 20
Concentration (%)
Figure 3. Precision vs. concentration for chromium in Ni-base superalloys.
miniature x-ray tube excitation noted ground and the instrument sensitivity. It
earlier. is found by three times the standard de-
Using the equation of the least- viation repeated measurements of the
squares curve fit through these data blank times the sensitivity, as deter-
points, we can extrapolate back to an mined by measurement(s) of a standard.
RSD of 33% to find the LOD. This value
agrees well with the LOD computed in References
the usual manner illustrated by equa- 1. IUPAC Compendium of Chemical Ter-
minology, 2nd Edition (1997). See
tions 1 and 2. Of course, this very time-
consuming means of computing the
website http://www.chemsoc.org/
chembytes/goldbook/.
LOD would rarely be used. Nevertheless 2. G.L. Long and J.D. Winefordner, Anal.
it is reassuring that this alternate method Chem. 55, 712A (1983).
produces the same result. 3. J.V. Gilfrich and L.S. Birks, Anal. Chem.
56, 77 (1984).
4. P.W.J.M. Boumans, Inductively Coupled
Conclusion Plasma Spectroscopy Part I, John
A picture is worth a thousand words Wiley & Sons (1987).
the diagrams presented in this tutorial 5. V. Thomsen, G. Roberts, and K. Burgess,
should help illuminate a sometimes Spectroscopy 15(1), 3336 (2000).
6. R. Payling, www.
confusing subject.
thespectroscopynet.com/
The limit of detection is related to the Educational/detection_limit.htm.
reproducibility of the spectral back-

Something
we should
know?
We welcome Write to:
your letters Spectroscopy
and appreciate 485 Route 1 South, Building F,
First Floor
your Iselin, NJ 08830
feedback. (732) 225-9500
Fax: (732)225-0211
E-mail: spec@spectroscopymag.com
Eugene, OR 97401-6806

114 Spectroscopy 18(12) December 2003

Circle 89 w w w. s p e c t r o s c o p y o n l i n e . c o m