Feasibility of locating tumours in lung via kinaesthetic feedback.

Abstract

Localizing lung tumours during minimally invasive surgery is difficult, since restricted access precludes manual palpation and pre-operative imaging cannot map directly to the intra-operative lung. This study analyses the force-sensing performance that would allow an instrumented kinaesthetic probe to localize tumours based on stiffness variations of the lung parenchyma.
Agar injected into ex vivo porcine lungs produced a model approximating commonly encountered tumours. Force-deformation data were collected from multiple sites at various palpation depths and velocities, before and after the tumours were injected.
Analysis showed an increase in force after the tumours were injected, in the range 0.07-0.16 N at 7 mm (p < 10(-4)). A 2 mm/s palpation velocity minimized exponential stress decay at constant depths, facilitating easier comparisons between measurements.
A sensing range of 0-2 N, with 0.01 N resolution, should allow a kinaesthetic palpation probe to resolve local tissue stiffness changes that suggest an underlying tumour.

Full text

Page 1

Feasibility of Locating Tumours in Lung via Kinesthetic Feedback
Greig L. McCreery
Ana Luisa Trejos
Michael D. Naish
Rajni V. Patel
Richard A. Malthaner
Abstract
Background
Localizing lung tumours during Minimally Invasive Surgery is difficult since restricted access pre-
cludes manual palpation, and pre-operative imaging cannot map directly to the intra-operative lung.
This study analyzes the force sensing performance that would allow an instrumented kinesthetic
probe to localize tumours based on stiffness variations of the lung parenchyma.
Methods
Agar injected into ex-vivo porcine lungs produced a model approximating commonly encountered
tumours. Force-deformation data were collected from multiple sites at various palpation depths
and velocities, before and after the tumours were injected.
Results
Analysis showed an increase in force after the tumours were injected, ranging from 0.07 to 0.16 N
at 7 mm, p < 10−4. A 2 mm/s palpation velocity minimized exponential stress decay at constant
depths, facilitating easier comparisons between measurements.
Conclusion
A sensing range of 0 to 1 N, with 0.01 N resolution should allow a kinesthetic palpation probe to
resolve local tissue stiffness changes that suggest an underlying tumour.
1Introduction
Minimally Invasive Surgery (MIS) is a surgical method which, due to its numerous advantages over
conventional open surgery, is being adapted to many surgical techniques. Some of the advantages
include a lower chance of infection, and reduced tissue trauma and post-operative pain [1]. Economic
advantage has also been cited as a benefit of MIS, due to reduced recovery time in hospital and a
faster return to work [2, 3]. These advantages are a result of the much smaller incisions, typically less
than 10 mm in length, that are necessary to gain access to the surgical site.
Irrespective of the advantages, some tasks in MIS are currently more difficult to perform than in
the corresponding open procedure, possibly resulting in longer operative time and thereby offsetting
some of the cost-savings and potential wait-time improvements. Some of the difficulties in MIS arise

Page 2

from reduced dexterity, poor ergonomic posture for the surgeon, lack of hand-eye coordination due to
the reversal of lateral movements caused by pivoting about the port of entry, and the inability of the
surgeon to directly manipulate the tissue [4]. Any interactions between the tissue and the instrument
are difficult to perceive due to friction and mechanical moments introduced at the trocar (a protective
sleeve inserted into the entry port) which tends to mask any tactile cues of the tissue interaction.
Advances in robot-assisted MIS have eliminated many of the potential drawbacks. The da Vinci?
Surgical Robot (Intuitive Surgical Inc.) dramatically improves dexterity and ergonomics through the
use of two 7 degree-of-freedom (DoF) articulating wrists which mimic motions at the surgeon’s control
console. This re-establishes direct hand-eye coordination by eliminating reversed tool motions and can
present additional advantages such as motion scaling and tremor filtering [5]. Nevertheless, the loss
of tactile cues remains a major drawback of MIS.
Tactile perception is an invaluable tool for many surgical procedures since it can provide rich
information on the mechanical properties of the tissue being manipulated. Since a malignant tumour
will typically be stiffer than the surrounding parenchyma [3], a surgeon can usually localize a tumour
via direct palpation when performing open surgery.
using an MIS method, the surgeon must locate the tumour without relying on direct palpation. The
common practice is to use standard MIS instruments to probe the surface of the lung, using any visual
or limited tactile cues to determine the position of the tumour. If available, MIS ultrasound can be
used. However, due to the poor image quality and artifacts caused by residual air in the lung, it is
not usually possible to find tumours smaller than 1 cm in diameter. If the tumour cannot be located,
the surgeon must increase the size of the incision and spread the ribs to allow finger access for direct
palpation.
Other imaging modalities are no better suited to locating a lung tumour intra-operatively. Pre-
operative Magnetic Resonance Imaging (MRI) or Computed Tomography (CT) imaging are used to
identify the region of the lung in which the tumour is located, but cannot be mapped directly to the
intra-operative site due to the deformability of the lung and its ability to move within the thoracic
cavity. Since these imaging modalities are not commonly available in the operating room, their use as
an intra-operative method of lung tumour localization is limited.
In view of these limitations, it is clear that an alternative method for locating tumours within the
lung parenchyma would improve the likelihood that a lung tumour resection can be completed using
minimally invasive techniques, thereby providing all of the associated benefits. One such method,
which has been the subject of considerable research, is the recreation of haptic cues, or the “sense of
touch,” from the tissue-instrument interaction to the surgeon-instrument interface. Haptic information
can be considered in two distinct modes: kinesthetic and tactile information [6].
Tactile information includes the sensation of surface textures, or distributed pressures acting across
the contacting surface. Measuring tactile information requires a tightly packed array of sensors capable
of measuring multiple contact pressures or forces concurrently. The use of tactile sensors to identify
pulmonary lesions was discussed in [7]. Validation tests using a foam model showed promising results.
Tactile feedback systems have also been proposed for identification and characterization of lesions in
the breast [8] and for identifying arteries during robotic surgery [9].
In contrast to tactile information, which requires a dense cluster of sensors, kinesthetic information
relates to the movement and bulk forces acting in the joints of an arm (human or mechanical) and
at the point of contact. Kinesthetic information may be used to assess the contour and stiffness
of an object and may be acquired using a simple force/torque sensor. Numerous researchers have
proposed using kinesthetic feedback to measure force and displacement during an MIS procedure for
various purposes. In [5], a 6 DoF force/torque sensor based on a strain gauge instrumented Stewart’s
Considering the resection of a lung tumour

Page 3

platform is proposed. The sensor is steam sterilizable and small enough to pass through a standard
trocar. Therefore, it can be located at the distal end of the instrument where trocar imposed friction
and moments will not interfere with the tissue interaction measurements. The sensor is designed for
use with a grasper system, and is capable of achieving a force resolution of 0.25 N in the z-direction and
0.05 N in the x- and y-directions. This sensor system is intended for use in a robotic surgical system
to help reduce unintentional damage to tissue and suture material by the inadvertent application of
excess force. A 3 DoF force sensor intended for the same purpose, but based on fiber-optic sensing, is
presented in [10]. The prototype exhibited a sensing resolution of 0.04 N.
A strain gauge sensorized laparoscopic grasper was developed in [11]. The grasping force and
grasper position were presented along with a measure of compliance, which could be used to differen-
tiate between objects of various stiffnesses. Another instrumented grasper, utilizing 2 thin foil strain
gauges, is described in [3]. This system is capable of operating in a wet saline environment due to
silicone encapsulation of the electronics, and can determine the location of the applied force along
the grasper jaws. The sensitivity can be adjusted by varying the amplifier gain, and the system was
reported to be sensitive to a force increase of “a few grams.” It was also shown through Finite Ele-
ment Analysis that the system could be used to measure distributed forces, approximating them as a
concentrated load. A computerized endoscopic surgical babcock grasper that utilizes existing surgical
tools is described in [12]. It performs an automatic palpation consisting of 3 cycles of a 1 Hz sinusoidal
displacement of the grasper. Experimental results indicating the tool’s ability to distinguish different
mechanical properties of tissues appear promising. In [13] tissue interaction was measured using a
number of strain gauges and a single-axis load cell integrated into a custom endoscopic instrument. A
different approach is used in [14], in which tissue stiffness is determined by measuring the amount of
current applied to the motor of a motorized grasper.
For the purposes of medical diagnosis, instrument design and improved simulations, in-vivo and
ex-vivo indentation testing of human and porcine abdominal organs was presented in [15]. It was
claimed that this included possibly the first in-vivo measures of tissue compliance from a human
subject. The results indicated that the stiffness between healthy liver tissue does not vary widely, but
that a diseased liver with an obstructed bile duct demonstrated a notable difference in compliance. It
was therefore suggested that an indentation testing instrument similar to that used in the study could
be used during MIS as a diagnostic tool.
To the best of our knowledge, the use of kinesthetic feedback from direct probing using a sensorized
instrument, rather than grasping, has not been applied to the task of tumour localization within the
lung. For this task, a probe instrument may be better suited than a grasper since it may be difficult
to reach all parts of the lung, or to fully capture the tumour when using a grasper instrument. It is
hypothesized that the force measured will be higher when probing tissue with an underlying tumour
than when probing intact tissue at the same depth.
The goal of this paper is threefold: 1) to determine with statistical confidence that the measure-
ments from indentation testing can indicate the presence of a tumour; 2) from the testing methods
used, to determine which performed the best in terms of measured force increase due to a tumour while
minimizing the number of false negative results; and 3) to determine the sensing range and resolution
that is required to realize the approach in actual MIS.
The design of a sensor system for accurately measuring the tissue-tool interactions in an MIS setting
presents considerable challenges in that the sensor must meet demanding size constraints, withstand
temperature variations, address issues of sterilization, and use biocompatible materials, while achieving
appropriate performance in terms of resolution and sensing range. While these constraints were not
addressed by the system used in these experiments, the results will be useful in providing target design

Page 4

specifications for systems which address such constraints.
2Methods
2.1The Model
In order to experimentally verify that kinesthetic feedback alone could indicate the presence of a
tumour in lung parenchyma, an accurate model was required. Ex-vivo porcine lung tissue was collected
from local abattoirs and used as a biological model for human lung. When not in use, the lungs were
individually sealed in plastic bags containing approximately 300 mL of saline and refrigerated. To
minimize the effects of autolysis of the tissue, all experiments were completed within 60 hours post-
mortem.
Several preliminary tests were undertaken to determine the most realistic method of simulating
tumours [16]. It was discovered that any method which required a major incision in the tissue changed
the force-deformation behaviour of the otherwise intact lung. Therefore, injecting material provided
the most realistic alternative. Sigma Gelrite Gellan Gum (agar) was prepared in a ratio of 225 mL
water to 7.5 g agar, boiled, and injected into the cold lung. Given that the tumours of interest were
approximately 1 cm in diameter, 0.5 mL of the agar preparation should have been injected. However,
initial tests revealed that the tumours formed were not perfectly spherical, and that some of the liquid
agar was lost through the injection puncture. Therefore, approximately 1.5 mL was used to form each
tumour. A sample of the artificial tumours, excised after testing was completed, is shown in Figure
1. No evidence of agar absorption into the surrounding parenchyma was noted during excision of the
tumours.
Experience in thoracic surgery has shown that the tumours typically encountered in lung can span
a range of stiffnesses. In qualitative terms, the feel of typical tumours commonly encountered in the
operating room can be described as ranging from the stiffness of a grape to that of a rock. The agar
tumours used herein fell within this range, as verified by an experienced thoracic surgeon.
In [17] it was shown that the Bulk Modulus and Young’s Modulus of canine lung are proportional
to the inflation pressure in the airways, being 4 times and 1.5 times higher, respectively. Currently,
during MIS lung tumour resection, the lung of interest is occluded at the bronchus and collapsed. Due
to the continued perfusion of the lung, most of the residual gases in the alveoli are absorbed by the
pulmonary circulation. The lungs tested in the following experiments were not pressurized. Due to the
elastic nature of the alveoli and terminal bronchi, it can be reasonably assumed the alveolar pressure
will be either equal to, or slightly greater than, atmospheric pressure. Hence, the stiffness of the
tissue will be either equal to, or slightly stiffer than that of an intra-operative lung. Since discerning a
tumour in stiffer parenchyma would be more difficult due to the less marked difference in stiffnesses,
the ex-vivo model used here represents a ‘worst-case’ model, compared to the intra-operative condition.
2.2Apparatus
The kinesthetic probe was realized using an aluminum rod approximately 50 cm long with a diameter of
9 mm and a hemispherical distal tip. A Gamma Force/Torque sensor from ATI Industrial Automation,
Inc. was placed in series between the proximal end of the probe and the mounting plate of a Mitsubishi
PA10-7C robot, Figure 2. The Gamma sensor was factory calibrated for 100 N measurement range in
the z-direction (axial direction of the probe), and 32 N in both the x- and y-directions. The resolution
is 0.003 N in the z-direction and 0.002 N in the x- and y-directions. Force data for the three orthogonal
vectors was used to calculate the magnitude of the resultant force vector, from which all of the analysis

Page 5

was conducted. It was assumed that using the resultant force vector would result in more robust data,
since slight deviations between the normal of the lung surface and the approach vector would not affect
the resultant force to the same degree as it would the axial force. Note that, for all tests conducted
in this study, the end-effector orientation relative the robot base was not modified and the probe was,
at all times, approximately perpendicular to the table on which the lung was tested. The effect of
various approach angles was not considered.
2.3Experimental Approach
Using kinesthetic feedback to measure a change in stiffness can be accomplished in two ways: The
tissue can be indented until a pre-set force is achieved, and the depth is recorded; or, the tissue is
indented to a specific depth from the surface, and the force is recorded. This question was examined
both experimentally and analytically in [18], where it was found that the constant depth scheme
is more sensitive to stiffness deviations when testing visco-elastic materials with an upward-sloping,
non-linear deformation curve. Therefore, the constant depth method was adopted for this study.
The lung sample was placed on a surgical towel which was clamped onto a fibreglass tray, thereby
preventing lateral movement of the lung during testing. The mounted sample was placed within the
robot workspace, with the probe aligned above the first palpation point.
The robot was programmed to move down, in a direction parallel to the long axis of the probe,
toward the surface of the lung in 0.5 mm increments until a threshold force of 0.04 N was detected
in the axial direction. This threshold was shown to be the smallest value that was not prone to
frequent false triggers due to the inertial forces of the probe. Once the threshold value was realized,
the robot was held stationary, during which time 17 measurements were recorded at a sampling rate
of 250 Hz. If the median of these measurements was greater than the 0.04 N threshold, the operator
was prompted to confirm that the probe was indeed in contact with the surface of the lung. This
redundant process was used to avoid “false starts” which occurred during preliminary tests that used
only a single measurement of the threshold value.
Once it was confirmed that the probe tip was just in contact with the surface of the tissue, a 10
second acquisition of continuous force measurements in three orthogonal directions was initiated at
a sampling rate of approximately 250 Hz. The probe was then advanced into the lung tissue to a
pre-programmed depth: either 5, 7 or 9 mm, after which it was held stationary for the remainder of
the 10 second sampling period. Once the measurement was complete, the probe was retracted to a
position above the lung, and moved to a point on the lung 30 mm away from any previous testing site
and the palpation process was repeated. Pre-planning of the testing grid ensured that only the lower
lobe was sampled, and that the main bronchial branches were avoided. After testing was completed
for each lung, a permanent marker was used to mark each palpation site on the surface of the lung.
Once all initial tests were completed for a specific depth, the agar was prepared and 1.5 mL were
injected into each palpation site, to approximately the mid-thickness of the lung. The palpation tests
were then repeated. Radiographs of the lung, such as that shown in Figure 3, revealed that since
the tumours occasionally did not form directly beneath the intended site, each tumour needed to
be located using manual palpation and aligned with the probe prior to testing. The test sites were
numbered such that the measurements from before and after the tumour was injected could be paired
for each location.
Before the data were analyzed, some measurements from either the test or control lungs were
removed because the operator had noted an error in the measurement process. Errors included in-
advertently testing extremely thin or damaged regions of the lung. Note that the data analysis was
blinded from the acquisition of the data, and any trials that were indicated to be questionable were

Page 6

removed without reviewing the results. After all questionable samples were removed, over 30 samples
for each depth remained.
Since the indentation velocity will affect the force-deformation behavior of a visco-elastic material,
the above mentioned experiments were performed with two variations of velocity control. In the
variable-velocity approach, the maximum velocity of the robot was left as the default of 40 mm/s.
Because of the short translations required in the palpation experiment, this velocity was not reached.
However, from preliminary experiments it was evident that the peak velocity increased with palpation
depth.Therefore, velocities could be considered equal when comparing uniform depths, but not
between depths. For the constant-velocity approach, the maximum velocity was set to 2 mm/s. Since
this velocity was attainable at all palpation depths tested, comparisons at constant velocity could be
made across all depths.
2.4 Pre-Conditioning and Experimental Controls
Over repetitive strain cycles, the response of soft tissue approaches a steady state stiffness and hys-
teresis. This effect is known as pre-conditioning [19], and is commonly employed in tissue research
to obtain more consistent results. In [19], in-vivo and in-situ tissue did not tend to reach a pre-
conditioned state within 10 cycles. Since tissue is not normally pre-conditioned during actual surgery,
the number of measurements taken at each point was minimized to reduce the pre-conditioning effect.
At each test site, only one sample was collected for the intact tissue, followed by a single sample after
the agar had been injected. Furthermore, test sites were placed at least 30 mm apart, thereby reducing
the influence from an adjacent test site. In total, 38 pairs of lungs were tested.
In [19], the differences in the force-deformation relationship between in-vivo and in-situ were
examined, indicating that the stress-relaxation and tissue recovery differed between the samples. Ad-
ditionally, [20] indicates that tissue properties can change by as much as 50% when comparing in-vivo
to ex-vivo samples, and that level of hydration, time post-mortem, temperature and tissue perfusion
all affect the visco-elastic properties of the tissue.
In order to isolate the changes that were caused by the presence of the artificial tumour from
changes caused by tissue autolysis1and other environmental factors, tissue controls were incorporated
into the testing. When the lungs were excised, the right and left lung remained connected by the intact
portion of the bronchi. In all cases, the left lung was used as the test lung with tumour injected, while
the right lung was used as the control. Both lungs were exposed to the same environmental conditions
and treatment, with the exception that no artificial tumours were injected into the control lungs.
Since the lungs came from the same animal and remained connected throughout the experiments,
the effect of factors such as time post-mortem, temperature cycles, hydration, etc., could be isolated
from the effect of the artificial tumour. Again, the test sites were numbered, such that the resulting
measurements at each location could be paired.
2.5Signal Conditioning
Continuous force measurements were collected for 10 seconds at a sampling rate of approximately
250 Hz. The raw data were post-processed in MATLAB using the filtfilt function and a second order
Butterworth filter with a cutoff frequency of approximately 2.9 Hz, as no high frequency signal was
expected. This filtering scheme filters the raw data in the forward direction, and then re-filters the
output in the reverse direction. This eliminates any phase distortion and also serves to effectively
1The enzymatic digestion of cells by enzymes present within them.

Page 7

double the order of the filter [21]. Figure 4 shows the raw and filtered data from two palpation tests
at 9 mm, using both the variable- and constant-velocity approach.
2.6Peak vs. Settled Measurements
Since lung tissue tends to exhibit a well known decaying stress with constant strain, as can be seen in
Figure 4, it is important that any comparisons between measurements be made at the same point on
the curve. Two points were considered in this study: the peak force measured, and the force after the
response had settled, taken at a constant number of data points from the peak. This was determined
separately for each of the six combinations of depth and velocity control approaches.
For each experimental approach, the number of data points from the peak to the end of the
measurement window was determined for each sample in the set, and the mean calculated. The
minimum of these values was set as the number of data points from the peak to where the settled
measurements would be read. However, since some samples exhibited an uncharacteristic peak near
the end of the measurement window, samples where this value deviated by more than 3 standard
deviations from the mean were permanently removed from further evaluation to avoid skewing the
read time of the settled value. To mitigate the influence of noise on the settled value, the median of
the 201 data points prior to the identified target measurement was reported as the settled value.
3Results
Each of the experiments outlined in Section 2 were performed in a paired manner, assessing the
response of the lung tissue before and after a tumour was introduced. Three different palpation
depths were considered with the probe advancing under both variable- and constant-velocity schemes.
These tests resulted in 6 data sets: 5 mm variable-velocity, 7 mm variable-velocity, 9 mm variable-
velocity, 5 mm constant-velocity, 7 mm constant-velocity and 9 mm constant-velocity. Futhermore,
the peak and settled force for each sample were determined following the methodology outlined in
Section 2.6, effectively doubling the number of data sets. In addition, each lung had an associated
control (in which agar was not injected), where each site was also tested twice, yielding an additional
12 data sets. The aggregate data from these 24 data sets was used to assess the maximum observed
forces and the observed force difference within paired palpations (pre- and post-tumour injection).
3.1Maximum Force Measurements
For the purpose of defining an appropriate sensor range for a kinesthetic lung tumour localization
system, it is necessary to look at the maximum forces occurring in each of the experimental approaches.
For this purpose, the peak measurement from each artificial tumour test, as determined in Section
2.6, is shown in Figure 5.
3.2Tissue Controls and Unadjusted Tumour Test Results
For both the artificial tumour tests described in Section 2.3 and control tests outlined in Section 2.4,
the change in force feedback given by ∆F = Ff− Fiwas determined for each pair of samples, where
Ffis the measurement recorded with the tumour present, or the second test of the control; and Fiis
the force measured during the initial palpation of either the test or control lung. These ∆F values
were grouped into sets according to depth, velocity scheme and peak or settled values. The results are
plotted as box plots and presented for the variable-velocity peak and settled, and constant-velocity

Page 8

peak and settled sets in Figures 6–9, respectively. Within each set, the tumour test data is adjacent
to the corresponding control data for each depth.
The upper and lower bound of the box plot represents the first and third quartile of the measure-
ments. The line through the box represents the median, while the notch represents the range of the
95% confidence interval of the median of the samples. If the notches of two sample sets do not overlap,
it can be concluded with 95% confidence that the true medians do differ. The whiskers represent the
range of measurements not including outliers, which are defined as measurements deviating from the
median by more than 1.5 times the interquartile range and are indicated by a ‘+’.
All p-values, determined using the Mann-Whitney Independent Sample Test, were less than 10−6.
This non-parametric test is analogous to the two-sample independent t-test, but is applicable when
the assumptions of normal distribution, implicit in the use of the two-sample t-test, cannot reasonably
be made [22]. All statistical analysis was performed using the MATLAB?Statistical Toolbox, and
verified using SPSS.
3.3Artificial Tumour Data, Adjusted for Controls
Before the results from the artificial tumour tests (see Section 2.3) can be interpreted, any changes
that can be attributed to factors other than the presence of a tumour must be accounted for. For
each test-control pair (see Section 2.4), the value of the upper quartile measurement of the control
was used to offset the tumour test data to account for these changes. The resulting tumour test
data, corrected for the controls, are shown for variable-velocity peak and settled and constant-velocity
peak and settled sets in Figures 10–13, respectively. Within each figure, the percentage of negative
differences (i.e., the second force measurement was less than the initial measurement) and the total
sample size is indicated for each sample set.
For this analysis, p-values were determined using the Wilcoxon Paired-Sample Test, and found to
be less than 10−4in all cases. This non-parametric test is analogous to the paired-sample t-test, but
again, is valid when assumptions of normal distribution and equal variance cannot be made.
4 Discussion
4.1 Control Validation
Examining the response of the controls in Figures 6–9, reveals that there was some change in the
force-deformation characteristics of the lungs which were not associated with the presence of the
artificial tumours. Factors such as temperature variations, tissue autolysis, hydration fluctuations,
tissue conditioning effects from the first palpation test, manipulation of the lung, or measurement
error may have contributed to these deviations. There is no discernable trend for this change across
the various palpation depths.
The p-values determined for the Mann-Whitney test provide a very positive statistic that the
tumour test results and control test results are different. This supports the methodology that assumes
there is some change in the force-deformation behaviour of the lung that can be directly attributed to
the artificial tumour.
By shifting the tumour test data sets toward zero by an amount equal to the upper quartile
measurement of the associated control response, the corrected data in Figures 10–13 provide a more
realistic indication of the force feedback deviations that can be expected in the presence of a tumour.

Page 9

4.2Statistical Analysis of the Proposed Method
It is clear from the corrected test results in Figures 10–13, that each experimental approach performed
reasonably well. The Wilcoxon Paired-Sample Test analysis indicates that regardless of velocity-
control schemes or measurement time, each approach produced a statistically significant change in the
force-deformation behaviour in the presence of a tumour, with all p-values being less than 10−4. This
provides a statistical indication that kinesthetic feedback alone could be used to detect these tumours
within lung tissue. However, some methods appeared to be consistently better than others.
The percentage of negative force differences provides an indication of the likelihood of arriving at
a false negative conclusion (i.e., not seeing a force increase in the presence of a tumour). Again, the 7
mm palpation tests consistently produced the least number of false negatives when testing the variable-
velocity approach. The 7 and 9 mm palpations both resulted in 3% negative force differences when
the peak force of the constant-velocity approach was analyzed, whereas the false negatives resulting
from the 7 mm palpation increased to 7% when considering the settled values. However, it should be
noted that the negative value causing this change was extremely close to zero, of the order −10−3. If
this value were assumed to be zero, the percentage of false negatives for the 7 mm test would again
fall to 3%. Furthermore, in an operative setting, multiple palpations by the surgeon may mitigate the
effect of false negative results.
Finally, the 7 and 9 mm constant-velocity palpation tests produced very similar performance in
terms of the median force difference measured, ranging from 0.12 to 0.13 N. While the medians were
very similar, the data spread tended to be tighter for the 7 mm tests and the false negatives were
a result of outlier measurements. The tighter data spread provides a more precise indication of the
force increase that would be associated with a tumour, and may allow a more accurate conclusion
regarding the presence of an underlying tumour. The greatest median ∆F overall was a result of the
9 mm variable-velocity test. However, this approach was also associated with the highest incidence of
false negative results.
The median force difference for the 5 mm palpation tests was lower than the 7 mm and 9 mm tests
regardless of measurement or analysis method. This indicates that while the 5 mm palpation tests
produced a reasonable test statistic, it was not optimal in any of the experimental approaches. It is
interesting to note that the data spread is minimal for all 5 mm tests compared to the other depths.
Presumably, the shallower palpation depth would not be influenced by a rigid support medium to the
same extent as deeper palpations. This is especially significant in thinner regions of the lung, where
deeper palpation depths might exhibit an altered response due to the underlying rigid support surface.
4.3Effect of Velocity
If the deviations observed with differing palpation velocity are due to the visco-elastic nature of the
tissue, it would follow that once the tissue had settled, there should be little difference between the
response of the variable- and constant-velocity tests, at the same palpation depth. By comparing
Figures 11 and 13, it is clear that the median of the settled responses for the variable- and constant-
velocity tests cannot be concluded to be different at a 95% confidence level. This suggests that any
difference in the measured peak response that is attributed to a higher palpation velocity, tends to
diminish as the sample settles.
Furthermore, comparing the peak and settled values between the two velocity schemes, it can be
seen that the 95% confidence interval of the median peak and settled values tend to have a greater
overlap when considering the constant-velocity (or slower) approach. This indicates that when using
the slower palpation approach, reasonably accurate comparisons could be made from readings taken

Page 10

after different settling times have elapsed. The nearly-constant stress resulting from the constant-
velocity approach shown in Figure 4 illustrates this well. This may be advantageous when considered
in a clinical setting, where maintaining a constant strain for an extended period may not be possible.
When considering the faster, variable-velocity approach, the pronounced peak and decaying strain
imply that to make comparisons between two tests, the time at which the response is analysed is
critical.
4.4Experimentally Determined Range and Resolution Required
The peak values shown in Figure 5 are the maximum forces that were recorded when testing each site
after a tumour had been injected. This gives a good indication of the maximum forces that can be
expected during the use of a kinesthetic instrument to locate a tumour in the lung.
With the exception of a few outliers, the peak measurements are less than 2 N and 1 N for the
variable-velocity and constant-velocity tests, respectively. Thus, it would be reasonable to define the
full scale range of a sensor system as 0 to 2 N; or 0 to 1 N, depending on the palpation velocity being
used.
The required resolution can be determined from the corrected ∆F values given in Figures 10–
13. Since the 7 mm constant-velocity approach appeared to perform slightly better than the 9 mm,
the values from this method, considering both the peak and settled analysis, will be considered in
determining the required resolution.
Considering the upper and lower quartile bounds from both the peak and settled analysis, a sig-
nificant portion of the ∆F values range from 0.07 to 0.16 N. Therefore, if a MIS kinesthetic sensing
system in unable to resolve force differences within this range, it would not likely be able to differ-
entiate the majority of these underlying tumours. Even a solution capable of resolving differences in
force as small as 0.07 N would result in false negative results in 25% of these tests. It should be noted
that in a clinical situation, multiple palpations around the suspected area would likely be employed.
This approach, combined with the currently available visual cues and limited tactile feedback trans-
mitted through the rigid instrument, may tend to reduce the likelihood of arriving at a false negative
conclusion.
In order to resolve the desired force differences, the resolution of the instrument should be one
order of magnitude less than this value, to avoid quantization error. Thus, a sensor resolution of 0.007
N, or approximately 0.01 N is suggested.
5 Conclusion
This study was performed to determine if there was a statistically significant difference in force-
deformation behaviour between intact lung tissue and tissue in which a lesion was present, using
ex-vivo porcine lung with injected agar simulating the tumours. The differences observed were used to
suggest an acceptable sensing range and resolution, and an appropriate palpation velocity, that may
allow a kinesthetic feedback instrument to be used to locate tumours in the lung during a minimally
invasive procedure.
Control tissues were used to quantify, and correct for, any change in the force-deformation behavior
of the tissue that may not have been due to the insertion of a tumour. The control lungs were from
the same animals and were exposed to the same environmental conditions and handling as the test
lungs, with the exception of the agar injection. Any statistically significant change that was observed
in the control was used to correct the tumour test measurements.

Page 11

Tests were conducted using a constant 2 mm/s palpation velocity across all palpation depths, as
well as using a variable-velocity approach in which the peak velocity increased with palpation depth.
The slower, constant-velocity approach appears to be superior since the exponential stress decay tends
to be significantly diminished compared to that in the variable-velocity approach. Thus, comparisons
can more accurately be made between measurements collected at any time after the desired palpation
depth has been reached.
Of the palpation depths tested, being 5, 7 and 9 mm, the 7 mm constant-velocity tests tended to
perform the best, in terms of statistical confidence in the observed force increase, the limited data
spread, and the number of false negative results. However, the performance of the 9 mm constant-
velocity approach was very similar to the 7 mm, with the exception of a wider data spread. Both the
7 and 9 mm constant-velocity approaches were clearly superior to the other methods tested.
It must be noted that even using the most successful approach, as much as 7% of the tests exhibited
no quantifiable increase in stiffness in the presence of a tumour, indicating that a false negative conclu-
sion would be inevitable in some cirumstances, regardless of resolution of the instrument. Nonetheless,
the use of a kinesthetic feedback system in a clinical setting could be enhanced by the limited visual
and tactile cues that are currently available, possibly reducing the likelihood of false negative conclu-
sions. Furthermore, multiple palpations of suspected sites would also serve to reduce the likelihood of
incorrect results.
The peak forces and differences in force-deformation behavior observed, and corrected for by the
control values, suggest that a full scale sensing range of 0 to 1 N would be appropriate for a kinesthetic
feedback instrument intended for use in lung tumour localization. To avoid quantization error, an
instrument with a minimum resolution of 0.01 N is required.
6 Future Work
The results from this study will assist in defining suitable operating characteristics for sensors used to
identify the location of lung tumours. Future work involves the development of clinical instruments
that use kinesthetic feedback for tumour localization in MIS.
7 Acknowledgments
The authors would like to thank the following individuals whose assistance was invaluable in the
realization of this study: Sheri VanLingen for her assistance in acquiring the test specimens, Dave
Browning and Dave Harrison for their help with the experimental set-up, Shiva Mohan for his support
with the data acquisition system, and Chris Kong for his assistance in the experimentation.
References
[1] Lanfranco AR, Castellanos AE, Desai JP, Meyers WC. Robotic Surgery: A Current Perspective.
Annals of Surgery. 2004;239(1):14–21.
[2] Cuschieri A, Berci G. Laparoscopic Biliar Surgery. Oxford, UK: Blackwell Scientific Publications;
1993.
[3] Dargahi J. An Integrated Force-Position Tactile Sensor for Improving Diagnostic and Therapeutic
Endoscopic Surgery. Bio-Medical Materials and Engineering. 2004;14(2):151–166.

Page 12

[4] Camarillo DB, Krummel TM, Salisbury JK. Robotic Technology in Surgery: Past, Present, and
Future. The American Journal of Surgery. 2004 Oct;188:2S–15S.
[5] Seibold U, Kubler B, Hirzinger G. Prototype of Instrument for Minimally Invasive Surgery with 6-
Axis Force Sensing Capability. In: Proceedings of the IEEE International Conference on Robotics
and Automation. Barcelona, Spain; 2005. p. 496–501.
[6] Ottermo MV, Ovstedal M, Lango T, Stavdahl O, Yavuz Y, Johansen TA, Marvik R. The Role
of Tactile Feedback in Laparoscopic Surgery. Surgical Laparoscopy, Endoscopy & Percutaneous
Techniques. 2006;16(6):390–400.
[7] Miller AP, Peine WJ, Son JS, Hammoud MD.
Nodules During Video Assisted Thoracoscopic Surgery. In: Proceedings of the IEEE International
Conference on Robotics and Automation. Roma, Italy; 2007. p. 2996–3001.
Tactile Imaging System for Localizing Lung
[8] Wellman PS, Howe RD. Extracting Features From Tactile Maps. In: Proceedings of the Sec-
ond International Conference on Medical Image Computing and Computer-Assisted Intervention.
Cambridge, UK: Springer-Verlag; 1999. p. 1133–1142.
[9] Beasley RA, Howe RD. Tactile Tracking of Arteries in Robotic Surgery. In: Proceedings of the
IEEE International Conference on Robotics and Automation. vol. 4. Washington, DC; 2002. p.
3801–3806.
[10] Peirs J, Clijnen J, Reynaerts D, Van Brussel H, Herijgers P, Corteville B, Boone S. A Micro
Optical Force Sensor for Force Feedback During Minimally Invasive Robotic Surgery. Sensors
and Actuators A: Physical. 2004;115:447–455.
[11] Bicchi A, Canepa G, De Rossi D, Iacconi P, Scillingo E. A Sensor-Based Minimally Invasive
Surgery Tool for Detecting Tissue Elastic Properties. In: Proceedings of the IEEE International
Conference on Robotics and Automation. vol. 1. Minneapolis, MN, USA; 1996. p. 884–888.
[12] Hannaford B, Trujillo J, Sinanan M, Moreyra M, Rosen J, Brown J, Leuschke R, MacFarlane
M. Computerized Endoscopic Surgical Grasper. In: Westwood JD, Hoffman HM, Stredney D,
Weghorst SJ, editors. Medicine Meets Virtual Reality 6. IOS Press; 1998. p. 265–271.
[13] Tavakoli M, Patel RV, Moallem M. Haptic Interaction in Robot-Assisted Endoscopic Surgery:
A Sensorized End-effector. International Journal of Medical Robotics and Computer Assisted
Surgery. 2005;1(2):53–63.
[14] Tholey G, Desai JP, Castellanos AE. Force Feedback Plays a Significant Role in Minimally
Invasive Surgery: Results and Analysis. Annals of Surgery. 2005;241(1):102–109.
[15] Carter FJ, Frank TG, Davies PJ, McLean D, Cuschieri A. Measurements and Modelling of the
Compliance of Human and Porcine Organs. Medical Image Analysis. 2001;5(4):231–236.
[16] McCreery GL, Trejos AL, Patel RV, Naish MD, Malthaner RA. Evaluation of Force Feedback
Required for Minimally Invasive Lung Tumour Localization. In: Proceedings of the IEEE/RSJ
International Conference on Intelligent Robots and Systems. San Diego, CA, USA; 2007. p. 883–
888.
[17] Lai-Fook SJ, Wilson TA, Hyatt RE, Rodarte JR. Elastic Constants of Inflated Lobes of Dog
Lungs. Journal of Applied Physiology. 1976;40(4):508–513.

Page 13

[18] Yen PL. Palpation Sensitivity Analysis of Exploring Hard Objects Under Soft Tissue. In: Pro-
ceedings of the IEEE/ASME International Conference on Advanced Intelligent Mechatronics.
vol. 2. Port Island, Kobe, Japan; 2003. p. 1102–1106.
[19] Brown JD, Rosen J, Kim YS, Chang L, Sinanan MN, Hannaford B. In-vivo and In-situ Compres-
sive Properties of Porcine Abdominal Soft Tissues. In: Westwood JD, Hoffman HM, Mogel GT,
Phillips R, Robb RA, Stredney D, editors. Medicine Meets Virtual Reality 11. IOS Press; 2003.
p. 26–32.
[20] Ottensmeyer MP, Kerdok AE, Howe RD, Dawson SL. The Effects of Testing Environment on the
Viscoelastic Properties of Soft Tissues. In: Proceedings of the Second International Symposium
on Medical Simulation. Cambridge, MA, USA: Springer; 2004. p. 67–76.
[21] Signal Processing Toolbox 6 User’s Guide. The Mathworks Inc.; 2007.
[22] Zar JH. Biostatistical Analysis. Upper Saddle River, NJ: Prentice-Hall, Inc.; 1999.

Page 14

List of Figures
1
2
3
A sample of the artificial tumours excised from the lungs after testing was completed.
The Mitsubishi PA10-7C palpating an ex-vivo porcine lung. . . . . . . . . . . . . . . .
Digital Radiograph of ex-vivo porcine lung after three agar simulated tumours were
injected. The three solid black lines in the left of the image are 22 gauge needles that
were inserted to mark the initial palpation site. A needle marker is also present in the
bronchus. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
The raw and filtered data of two 9 mm palpations: one taken from the variable-velocity
scheme, and the other from the constant velocity scheme. . . . . . . . . . . . . . . . .
Peaks of all force measurements recorded with a tumour present. Both constant- and
variable-velocity results are shown. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
∆F of the peak measurements from the variable-velocity experimental approach. . . .
∆F of the settled measurements from the variable-velocity experimental approach. . .
∆F of the peak measurements from the constant-velocity experimental approach. . . .
∆F of the settled measurements from the constant-velocity experimental approach. . .
∆F of the peak measurements from the variable-velocity experimental approach, cor-
rected for the deviations in the associated control.
∆F of the settled measurements from the variable-velocity experimental approach, cor-
rected for the deviations in the associated control.
∆F of the peak measurements from the constant-velocity experimental approach, cor-
rected for the deviations in the associated control.
∆F of the settled measurements from the constant-velocity experimental approach,
corrected for the deviations in the associated control. . . . . . . . . . . . . . . . . . . .
15
16
17
4
18
5
19
20
21
22
23
6
7
8
9
10
. . . . . . . . . . . . . . . . . . . .24
11
. . . . . . . . . . . . . . . . . . . . 25
12
. . . . . . . . . . . . . . . . . . . .26
13
27

Page 15

Figure 1: A sample of the artificial tumours excised from the lungs after testing was completed.