Biomedical Engineering Reference
In-Depth Information
receiver operating characteristics (ROC) curves. In the
skin context, reasonable clinical requirements on area
under ROC curve have been published by Lee and
Claridge (2005). A number of classifiers have been tried
on data sets prepared by data decomposition techniques
using indices as well as PCA, and using impedance data
from both surface electrodes and spiked electrodes, by
˚
berg et al. (2003b, 2004, 2005). It seems that non-mel-
anoma skin cancers (NMSCs), which are slow growing,
degrade the skin barrier enough to make it more conduc-
tiveandinthiscasetherewouldbenoneedfordemolition
or short-circuiting of the
stratum corneum,
while, for ex-
ample, malignant melanoma at an early stage the
stratum
corneum
can be quite normal (electrically insulating) and in
this case the spiked electrodes increase diagnostic power.
Figure 4.1-25
illustrates a PCA plot of NMSC and
other lesions (neither normal skin nor malignant mela-
noma) using two principal components which in this case
describe 84% of the variation in the data set that was
obtained with non-invasive probes (without spikes). It is
obvious that different ways to cut out the data volume
including the cancer lesions would include different
amounts of non-cancer data points, and that it is generally
impossible to avoid classification of at least some harm-
less lesions as harmful. The choice of classifier and cut-
off levels is a matter of risk assessment, common sense
and data characteristics. In the case of malignant mela-
noma, missing a tumor might be assigned an unacceptable
risk (life at stake), and therefore an almost 100% sensi-
tivity required, which would entail a reduced specificity
at a given area under ROC curve. In other cases it might
be more important with high specificity to avoid painful
and costly interventions, because a wrong diagnosis is not
dangerous and the patient simply could come back if the
complaint remains.
An update on this group's latest work on validation of
their skin cancer detector in multi center clinical studies
can be found in Ollmar et al. (2007).
Subcutaneous fibrosis is a common side effect of ra-
diotherapy given, for example, to women with breast
cancer. Nuutinen et al. (1998) measured the relative
permittivity of the skin at 300 MHz with an open-ended
coaxial probe, and found that the permittivity values
were higher in fibrotic skin sites than in normal skin.
Based on in vitro experiments with protein-water solu-
tions indicating that the slope of the dielectric constant
versus the electromagnetic frequency is a measure of the
protein concentration, Lahtinen et al. (1999) demon-
stratedthatskinfibrosiscanalsobemeasuredwiththe
slope technique. Both Nuutinen et al. (1998) and Lahtinen
et al. (1999) found a significant correlation between the
permittivity parameters and clinical score of subcutaneous
fibrosis obtained by palpation. Finally, radiation-induced
changes in the dielectric properties were also found in
subcutaneous fat by modeling the skin as a three-layer
dielectric structure (Alanen et al., 1998).
4.1.15.4 Electrodermal response
The sweat activity on palmar and plantar skin sites is very
sensitive to psychological stimuli or conditions. One will
however usually not be able to perceive these changes in
sweat activity as a feeling of changes in skin hydration,
except, for example, in stressing situations like speaking
to a large audience. The changes are easily detected by
means of electrical measurements, however, and since the
sweat ducts are predominantly resistive, a low frequency
conductance measurement is appropriate (Grimnes, 1982).
Electrodermal response (EDR) measurements have during
many years been based on DC voltage or current, and ac-
cordingly the method has been termed galvanic skin re-
sponse (GSR).
The measured activity can be characterized as exoso-
matic or endosomatic. The
exosomatic
measurements are
usually conducted as resistance or conductance mea-
surements at DC or low frequency AC. Resistance and
conductance will of course be inverse when using DC
excitation, but when AC excitation is used, it is impor-
tant to remember that resistance generally is part of
a series equivalent of a resistor and a capacitor, while
conductance is part of a parallel equivalent of these
component. In this case it is obvious that resistance and
conductance are no longer inverse, and conductance
should be preferred to resistance since ionic conduction
NMSC (
n =
198)
Other lesions (
n =
1231)
PC1 (71.6%)
Figure 4.1-25 PCA plot of NMSC.
