Hand-Impedance Measurements with Robots during Laparoscopy Training

—This paper presents hand-impedance measure- ments during laparoscopic training with physically interactive manipulators. We develop a co-manipulated robotic system allowing hand-impedance measurements in an active manipulation task with occasional environmental contact. Six professional, four trainee surgeons, and ten novice subjects participated in our experimental program for a suturing activity where the novice subjects were involved in a ﬁve weeks training practice. Variable admittance controlled robots, attached to the tools with force sensors, applied step vice velocity disturbances while subjects were trying to set the needle perpendicular to the surgical driver. Hereby, impedances of the left and right hands were computed in four different directions. Then, the measured impedance parameters across all subjects were compared with respect to the participants’ level of proﬁciency and skill progres- sion via statistical analyses to demonstrate effectiveness of the system. Results indicate that hand-impedance in the direction of the suturing-line demonstrates a consistent change throughout training and across different levels of expertise in laparoscopy. Therefore, hand-impedance information, proposed here, can pave the way for future development of robotic assessment or assistance in laparoscopy training programs.


I. INTRODUCTION
Minimally Invasive Surgery (MIS) methods are aimed at reducing the damage to the body tissues during diagnostic or surgical procedures, which connotes less post-operative pain, lower risk of infection, and a quicker recovery time for the patients as compared to conventional (open) surgery [1]. Laparoscopy is minimally-invasive inspection and surgery inside the abdominal cavity; surgeon can access inside of the abdomen or pelvis with minimal surgical wounds by using laparoscopic instruments such as small-scale tubes and cameras (known as endoscope). Improvements within the scaled-down display devices and special surgical instruments have given rise to the utilization of this technique. Consequently, laparoscopy has become the main method for surgical procedures around the abdominal region such as cholecystectomy and appendectomy surgeries [2], [3]. In a more advanced robotic system surgeons operate robots from a console based on the 3D image via two master controllers in this way robots can enhance motion precision (e.g, via suppression hand vibrations). The focus of this paper is laparoscopy (manual) training, and not robotic surgery [4]; This  Laparoscopic technique also brings additional challenges to the surgeons: operations are difficult to learn and perform. The primary challenges in laparoscopic procedures include the disturbed observation through a non-stationary camera platform and the loss of depth perception as the operation is viewed on a two-dimensional flat screen [4], [5]. Besides, manipulation is non-intuitive due to the discrepancy between the hand movements and tip of the laparoscopic tools; the well known fulcrum effect. The usage of these tools also leads to the loss of tactile sensing [6]. To cope with these difficulties surgeons are required to carry out an extensive training program, where with limited one-to-one expert guidance trainees try to learn from their own mistakes or through the feedback of virtual trainers based on the count of some task related performance measures [7], [8].
In laparoscopy training, coaching has been proven to have significant influence on the learning curve, thereby structured coaching not only with expert surgeons but also with assistive robots might present a key element in the acquisition of the laparoscopic surgical skills [9], [10]. Hand impedancemeasurements can potentially be used for both assessment and training purposes for laparoscopy. The current paper focuses on the assessment aspect, by making measurements and using them to distinguish between novice and professional performances. However the impedance-measurements have the potential to be used to provide feedback on their hand-impedance characteristics and to inform them how to change the use of hand/body [11] to bring the handimpedance to an optimal, similar to what expert coaches do. The current paper presents a first study to measure handimpedance in laparoscopy activity, and therefore is a first step towards exploring the promises of such measure which constitutes a biomechanical measure of performance. Currently, the assessment techniques for laparoscopy mostly focus on the kinematic movement of the tools/surgeons' upper limps (short distance, time, frequency content, etc.) [12], [13], [14]. There is no criterion to our knowledge that monitors the state of human biomechanics, such as hand-impedance or muscle activity, and no objective assessment method that monitors directly the biomechanical behaviour of the trainee. A biomechanical measure of performance might provide a an objective assessment, difficult, if not impossible, to trick. Our study, in this sense, provides a novel measure with a totally new characteristics, a measure of biomechanical behaviour, inspired by our previous work on manual welding [15], [16].
In this paper, whose preliminary findings involving only novice subjects with a short period of time training were presented in [17], hand-impedances of six professional and four trainee surgeons along with ten novice subjects, who participated in a five weeks laparoscopy training program, were measured during laparoscopy suturing experiments. Measurements were performed while participants were trying to set the needle with respect to the needle driver for preparation to enter the suturing pad. Small step-vice velocity disturbances were applied with the robotic manipulators while participants were manipulating the needle. Various impedance parameters measured in different directions at different periods within the program and statistical analyses were carried out to identify any significant difference between the expertise levels.
II. RELATED WORK The musculoskeletal system forming the human hand along with the arm can be associated or assumed to behave, all together, as a mechanical system. Dynamical characteristics of this system, in general, is described as a mechanical impedance while excluding the voluntary conducts [18]. This impedance characteristic is encountered frequently within the human-robot interaction such that any slight vibration or oscillation at the point of touch involuntary extinguishes with a human hand grip [19]. In addition to this passive behaviour, human hand reaction to external disturbances is modelled locally as a Linear Time Invariant (LTI) system consisting of a mass, spring, and damper [20].
Based on such modelling, measuring human handimpedance or arm joint impedances implies estimation of the mass, spring, and damping parameters within the aforementioned LTI model. Applying small impulse type force or position perturbations from a grip point and analysing the resulting response behaviour of the hand, such as interaction force and displacement from the equilibrium posture, is extensively revisited methodology; for instance see [21], [15] for the force and [22], [23], [24] for the position disturbances.
To apply perturbations and measure the hand position, admittance controlled robotic manipulators have been used in our previous studies [15], [25] within the aforementioned techniques without considering the overall system's stability. This technique allows us to measure the hand-impedance, during actual and professional task execution, affected by the muscle activity levels and hand-arm orientations specific to the professional task. In this way, hand-impedance measurements can be applied to real-life problems [16], rather than only to the laboratory devised experimental manipulation and only for passive (inactive) arm impedance measurements. But, unlike in applications such as manual welding [15] or airbrush painting [25], the training for suturing requires frequent contact of the laparoscopy instrument with hard (key hole, the needle and tip of the other instrument) and soft (the pad to be sutured) structures. Thus, special care must be taken to eliminate the instability that might occur due to such contacts during the training and measurement procedures.
To assess the stability of an interactive robotic architecture, passivity, a sufficient condition for the stability, is the main technique applied by many researchers. This method provides an elegant tool to eliminate severe constraints caused by the unmodelled dynamics of the robotic systems by considering only the input and output energy [26]. But, a major problem with this method is that the overall design becomes too conservative. To reduce the conservatism, one can design passivity observer/controller [27], by using measured forces and velocities to estimate the total power or the energy injected to the system. Yet, integrated energy during passive motion is the inherent limitation of this method; that phenomenon prevents instant active behaviour detection in real time implementation and requires intuitive energy resetting methodology [28], [29].
Recently, attention has also been focused on deriving empirical instability detection methodology by analysing forces or motions of the robot in the frequency domain. This procedure is intuitively stating that a stable motion does not exhibit unintentional high frequency movements or vibrations. By distinguishing the desired motions from the undesired ones via haptic stability observer (HSO) [30], unwanted actions can be eliminated via penalization techniques such as an increase in the overall impedance of the system by appropriate control action [30], [31], [32]. Here, following [30] and [32], we implemented an adaptive admittance control that allows both transparent co-manipulation in normal manipulation conditions and low admittance in case of oscillations during contact.
Accordingly, we developed a robotic measurement system for hand-impedance measurements in an active manipulation task with occasional environmental contact, implementing the system to the case of laparoscopy training exercises. Measured parameters, here, are comparable to and in the same order as in previous hand-impedance measurements (see, e.g., [15]). Statistically significant difference was observed in some parameters across the surgeons, also before and after training of the novice subjects and plausible implication of this was proposed. This observation would support the idea that impedance measurements relate to laparoscopic manipulation skills that are gained through training and experience.

A. Experimental Setup
The experimental setup consisted of an MIS training box, 2 Universal Robots (UR3), and 2 ATI Gamma force/torque (FT) sensors (with ATI FT 9105-NETB sensor box). The FT sensors were inserted between a special mechanical adaptor which was integrated to the MIS tool and the robot's endeffector, see Fig. 1. The UR3 robots are lightweight, have six degrees of freedom, capable of carrying 3 kg at their end effectors, and controlled by their own control boxes providing 125 Hz control cycle. To create a human-robot interaction, an admittance control architecture with variable parameters was implemented by using the Robot Operation System (ROS) and the FT sensors' measurements with a sampling frequency (f s ) equivalent to the robot's control cycle (f s = 125 Hz).

B. Experimental Procedure: Setting of the Needle
Needle setting was chosen as a target task because setting the needle perpendicular to the bar of the needle driver is one of the most difficult steps and perhaps the most crucial step in an effective laparoscopic suturing. While the subjects were setting the needle robotic manipulators connected to the two MIS tools, the receiver (left hand tool) and the driver (right hand tool), passively followed the hand movements and they became active from time to time to introduce slight disturbances for the measurements. In this study we followed the needle setting instructions and procedure as described in [33]. The experiment starts by placing the needle at the right half side of the suturing rag as illustrated in Fig. 2 (a). The participants were instructed to set the needle as shown in Fig. 2 (b). The subjects were advised to hold the front section of the needle with the receiver and back section of the needle with the driver, thus the unnecessary steps were eliminated to reduce the task completion time, which would eventually reduce the effect of the fatigue, see Fig. 3 (a) for the graphical illustration of the mentioned needle segregation. Additionally, in order to clarify the step-wise "needle dancing" technique (i.e. positioning the needle and orienting the angle), the needle was hypothetically divided into three sections as illustrated in Fig. 3 (b).
Then, the subjects were instructed to follow the subsequent steps to set the needle; • A→Drag: The driver starts at position 'A', see Fig.3 Then, passes the needle to receiver which will hold the needle at position 'B'.
• B→Right: The needle is righted to correct the position by pulling/pushing the driver from point 'A' and rotating the receiver from point 'B'. Once the needle is at the right angle the driver is released from point 'A'.
• C→Confirm: The needle is now grasped by driver at point 'C' and locked. The orientation is tested with an axial rotation of the driver. If the angle is not appropriate for entering the skin, then the process is repeated in the reverse order. We refer the reader to [33] for more detailed information and for other advanced needle setting techniques.
Before the experiments, the participants were introduced to the MIS training kit and usage of the MIS tools, they then were instructed about the process of suturing. The overall experimental protocol was as follows; 1) Participant grasped the handles. The needle was initially stationary on the pad as in Fig. 2 (a), 2) Participant picked the needle's strand with the driver (first part of the step A), 3) Participant passed the needle to the receiver (second part of the step A), 4) Participant corrected orientation of the needle (step B), 5) Participant re-grasped the needle with the driver at the correct location and tested the orientation (step C), 6) Participant repeated steps 3-5 until a successful grasp/orientation by the driver was achieved, 7) Participant repeated steps 1-6 within an experimental session (max 25 min). For the novice participants, a demonstration of a complete suturing with and without robots were performed by the authors to demonstrate these in practice.

C. Empirical Instability Disclosure
Stability is inherently the main concern while designing a control architecture for a robotic system interacting with its environment. Therefore, there has been a great deal of effort to design absolutely stable interactive robotic systems whose application areas vary from industrial and military to bilateral teleoperation [34], [35], [36], [37].
Inspired from [30] and [32], a variable admittance controller was designed by using the HSO index, I p , and a recursive stability index, I f . Those indices can be determined by analysing the robot's velocities/positions or interaction forces in frequency domain via using FFT. A ratio, known as HSO, is obtained by dividing the sum of the amplitudes of unstable frequency components with the sum of the amplitudes of all frequency components. The changes in this parameter can be used as a remark to detect the overall instability (or stability).
where P f (f ) of the frequency components f can be calculated via FFT of the determined signal (e.g., force) and f 0 denotes the lowest frequency within the FFT. Another, more commonly applied, recursive stability index is given as where I f rms is the ratio between the root mean square and the maximum value of the measured force signal and λ is a tunable time constant of the index [32]. Ultimately, one can enhance system's robustness by associating increase of I f to an overall impedance increment in the virtual end-effector dynamics of the admittance controlled system to empirically provide stability.

D. Implemented Admittance Control Architecture
The admittance control architecture's block diagram is given in Fig. 4. The force sensor at the robot's end effector measures the interaction force with the tool kit and based on this measurement the controller generates desired velocities. In Cartesian space, the motion dynamics of the admittance controller can be described as where F s , V ref ∈ R 6 denote the measured interaction force/torque and desired end effector velocity vectors. The diagonal matrices M a , D a ∈ R 6×6 are controller's virtual mass and damping, respectively. The desired velocities in Cartesian space, given in (1) and computed based on the interaction force/torque along with the virtual mass and damping, can be transformed into the joint space by using the robot's Jacobian matrix J(q) ∈ R 6×6 . One can determine the desired robot joints' velocities,q ref ∈ R 6 , while assuming that the inverse of the Jacobian matrix exists (robot is not operating nearby the singular joint configuration) aṡ

Sensor
Force Torque Admittance control parameters, M a and D a , need to be meticulously determined due to the inherent trade off between the stability and transparency of the robot while following the human hands' movements [38]. Here, the controller's parameters, mass and damping, were designed to be variable parameters and their alteration was associated with the stability indices as where m min and d min are minimum virtual mass and damping parameters such that stable free space movement is maintained, I ∈ R 6×6 denotes the identity matrix, and D u ∈ R 6×6 is used to express the dimensionless quantity I f in physical units and it was denoted as D u = d min I Ns/m. When an oscillation is detected the variable parameters, in (2), increase based on the increments in the stability index. Thus, the unwanted movements leading to the instable behaviour are suppressed by increasing the overall impedance of the robot end-effector.

E. Frequency Analysis of the Laparoscopic Operation with Robots
Dynamics of the intrinsic hand movement in daily use are significant over the low frequency range, 0-10 Hz [39]. Laparoscopic operations require gradual and dedicated hand motions; therefore one can expect to have significant frequency content during the operations in a much lower bandwidth than 0-10 Hz. To quantitatively determine principal frequencies during laparoscopy, initially, we carried out different experimental scenarios where laparoscopic tools were manipulated by the authors first in a gradual, stable manner and then in a fast and oscillatory manner, which can be characterized as an undesired, instable movement. The stable motions were achieved under high mass and damping parameters (m = 5 − 7kg, d = 50 − 100Ns/m) within the designed admittance controller, similarly instability was obtained under low admittance control parameters (m = 0.5 − 2kg, d = 5 − 30Ns/m). After running 6 different experimental scenarios for each of stable and unstable motions (12 in total), we have analysed the interaction forces and Cartesian space velocity measurements in frequency domain by using fast Fourier transform (FFT) [40]. The frequency spectrum of the both signals are illustrated in Fig. 5.
As seen from the frequency spectrum, the principal frequencies of the desired stable motions in both measures (force and velocity) are lower than 1 Hz. On the contrary undesired, instable, motions' principal frequencies are settled at frequencies higher than 2 Hz. In this regard, a finer frequency resolution, f ∆ , within the FFT analysis enables us to distinguish the principle frequencies of the desired and undesired motions. To obtain this, a large value of FFT window size, N , with respect to determined sampling frequency, needs to be chosen, as f ∆ = fs N . Based on this, we have used the frequency resolution f ∆ = 0.9766 Hz, by choosing N = 128, and identified the critical frequency, f c , as 1.9531 Hz in order to distinguish between stable movements and unstableinvoluntary-oscillatory motions. In this way, any movements higher than the critical frequency were interpreted as an involuntary behaviour. Besides that, we chose to use the force signal (as in [32]) in the forthcoming frequency analyses as force becomes the dominant measure when the MIS tool is in contact with its environment, typically while suturing.

F. Subjects and Impedance Measurements
Six consultant surgeons (male), experienced in laparoscopy, and four trainee surgeons (male), experienced in traditional surgical procedure yet in a training program for laparoscopic operations, voluntarily participated the experiments in the Cuschieri Skills Centre at the University of Dundee. In average, the professional surgeons had 15 years' expertise in general surgical practice and 133 hours in laparoscopic operations, and the trainee surgeons had 5 years' surgical expertise and had gained 3 hours laparoscopic training. Based on the test results of the Edinburgh Handedness Inventory [41], one of the professional and one of the trainee surgeons were left-handed and the rest were righthanded.
Additionally, ten novice subjects (5 males and 5 females) took part in a five weeks training program, where experiments and measurements of 6 hours in total per participant took place in Week 1 (W1), Week 3 (W3), and Week 5 (W5). Week 2 and Week 4 were considered to be training only slots where trainees practised the exercises (4 hours in total per participant) so no measurements were taken during these periods. All the novice subjects were recruited among the PhD students of the Institute of Sensors, Signals, and Systems at Heriot-Watt University (HWU), on a voluntary basis. The novice subjects did not have any prior experience on laparoscopic operations and they used the MIS training kit for the first time during our experiments. All the novice participants considered themselves as right-handed, yet according to the Edinburgh Handedness Inventory two of them were actually mixed-handers. The experiment protocol was approved by the Ethics Committee of the HWU. All the participants were provided with an information sheet, and they gave their informed consent prior to the experiments. The participants were instructed about how to set the needle, enter the skin with a needle, and tie two different surgical knots by using MIS tools. In the beginning, the subjects familiarized themselves with the MIS training system. They performed setting the needle process freely without the robots. After a couple of successful attempts, the subjects carried out the same process while adaptive admittance controlled robots with m min = 5 kg and d min = 50 Ns/m were attached to the tools and passively followed the hand movement of the participants.
During the measurement experiments, the subjects performed the needle setting process/task inclusive of the disturbances, as seen in Fig. 6. The subjects were informed in advance that the robots would apply perturbations. To prevent any voluntary actions against the disturbances, the perturbations were composed of 100-ms duration 0.15 m/s velocity impulses (v dist ) in one of the eight (±v n , ±x, ±y, ±z) directions, randomly applied without replacement as, where t d denotes moment before the disturbance and t u − t d = 100-ms [42]. Fig. 7 illustrates one of the implemented perturbations during the experiment. Here, v n corresponds to the direction perpendicular to the moving plane of the tools (as explained in the subsequent paragraph), x corresponds to the direction that the subjects are faced and which is along the suturing line, y corresponds to the direction perpendicular to the subjects and the suturing line, and z corresponds to the direction parallel to the gravity, see the x and y directions illustrated in Fig. 1. The disturbances were introduced at random instances, making sure that there were at least 4 sec in between. Also to reduce the effects of the fatigue, particularly with the novice subjects, each experimental session lasted around 18-25 min [43], [44]. The aforementioned system's parameters used within the experiments are given also in Table I.   Fig. 7: Data collected from one of the experiments and exclusively zoomed into the disturbance period for the illustration and clarity. The left column shows when a disturbance was applied in (vn) direction and right column shows when disturbance was applied in x, y, and z direction. Mismatches between the actual (p) and predicted (p) positions are illustrated in the second row to indicate the effect of the perturbations. Third and the last rows illustrate the measured velocities and forces, respectively. ing direction, v n , besides the globally fixed Cartesian directions, was motivated by the idea that humans might be modulating the directional impedance according to the movement of the hand. To define a motion plane regarding to the action of the robot's end effector at the Cartesian space, three consecutive tool positions were chosen as P 0 = (p x0 , p y0 , p z0 ), P 1 = (p x1 , p y1 , p z1 ), and P 2 = (p x2 , p y2 , p z2 ) such that P 0 denotes the most recent position estimation attained via designed Kalman filter based on the dynamic model related to position (p k ) and velocity (v k ) as wherep,v, and ∆t denote estimated position, estimated velocity, and the sampling time, respectively. The motion plane can be represented with two vectors − − → R 10 = (p x0 − p x1 , p y0 − p y1 , p z0 − p z1 ) and − − → R 12 = (p x2 − p x1 , p y2 − p y1 , p z2 − p z1 ) which are denoted by the tool's positions. Thereafter, the normal vector (v n ) can be determined by the cross-product of these two vectors ( − − → R 10 × − − → R 12 ) which is inherently perpendicular to the motion plane defined by the end effector positions. A predefined amplitude can be associated to |v n | by normalizing the vector. A graphical illustration for this type of vector on the Cartesian plane is given in Fig. 8.  The use of ranking scales (knot security, symmetry of suture, position of suture, operative times, etc.), recording the penalties and mistakes made during the laparoscopic procedure is commonly used assessment technique [45]. Also, image processing or accelerometer data analysing techniques, which require additional effort to create algorithms, are applied to assess motion quality and smoothness during laparoscopic operations [46], [47], [48]. However, the methods reported are not yet standardized, availability is limited, and the process involves complex recording systems and image analysis [49]. To quantitatively illustrate the expertise J o u r n a l P r e -p r o o f Journal Pre-proof level of the surgeons and progression of the novice subjects throughout the training program we have simply counted how many times they have completed the needle setting task (T ) during the experiments. Number of the needle dropping (P ) was used as a penalization criterion, thus any unnecessary movements extending duration of the tests were avoided by the participants during the experiments. Subsequently, an overall performance per minutes (opm) criterion was calculated by subtracting number of the penalties from the total number of the completed task (T − P ) and dividing this with the time taken (min) by individual subjects during the experiments. In this way we constructed a practical and easy measure of performance for our purpose of assessing the specific exercise we employ, through capturing the three basic criteria; achievement, operation time, and the mistake of dropping the needle.

G. Hand Impedance Estimation
Human hand impedance was modelled as an LTI passive operator in each of the three main directions (x, y, z) and also perpendicular to the moving direction (v n ) as where m h , d h , and k h are the mass, damping, and stiffness parameters of the human hand contact impedance in Cartesian and v n directions and ∆p states the position of the hand in Cartesian space. We perform decoupled measurements as we apply single disturbance in one of these directions and measure the reaction in the same direction. By using measured/calculated data of force (∆f ), position (∆p), velocity (∆ṗ), and acceleration 1 (∆p), the equality in (3) can be solved for estimating the impedance parameters by using the well-known ordinary least-squares method as where X state = [∆p ∆ṗ ∆p] and means transpose. In addition to that, we also estimated rate-hardness (rh) measures in each specified directions as proposed in [50], as a more intuitive measure of the human-hand resistance to external disturbances. The reader is refered to [15] for more detailed information about how to calculate the rh, ∆f , and ∆p values in Cartesian coordinate directions. Briefly, voluntary motion along the axis of perturbation is eliminated from our computation by subtracting the velocity and force at the instant just before the disturbance from the velocity and force recorded after the disturbance. In this way, only the velocity and force components that result from the disturbance are retained and the impedance values are computed with these retained trajectories. Differently, to estimate the relative interaction force and displacement in v n direction we have used the projection of these two measures, for instance the relative interaction force in v n direction can be estimated as ∆f = |∆f x | cos(φ x ) + |∆f y | cos(φ y ) + |∆f z | cos(φ z ), 1 Obtained numerically via implementing finite difference approximations to the filtered velocity measurements with, where f i and φ i correspond the measured force in a specified direction and the angle between the nominal vector and the coordinate directions, respectively (i = x, y, z), see Fig. 9 for a sample of measurement with estimated signals. Analogues calculations were carried out to estimate the relative displacement in v n direction as well.  In addition to the hand-impedance measurements, we also investigate whether there exists any statistically significant trend of change at hand-impedance throughout laparoscopy training and difference between the groups that have been established based on the expertise levels of the subjects.

IV. MAIN RESULTS
The means and standard deviations (σ) of the left and right hands' measured impedances in all the directions are given in Table II and rate-hardness values are also illustrated in Fig. 10 for easy comparison. To determine whether there exists any meaningful difference between the impedance measurements among the weeks and expertise levels, we analysed statistical significance of all the estimated data groups (128 in total).
Before the analyses, we first removed the excessive outliers from the data groups by eliminating any measurements that were above (Mean + 5σ) and any measurements below (Mean−5σ) values [51], then normality tests were carried out for the data groups and we applied Box-Cox transformation 2 to the groups failed within the initial test in order to achieve a normalized distribution in each of the compared groups. Approximately, 94.5% of all the groups passed either the  Lilliefors or the Anderson-Darling normality test with a significance level p = 0.05 and the data sample size, that executed within the normality test, was > 100. The data groups that failed in the normality tests, despite the transformations, were graphically inspected (via Histograms and Quantile-Quantile Plots) and outliers that jeopardize the normality were ignored in the forthcoming analyses. We applied Welch's t-test (Matlab ttest2()) to analyse the impact of expertise with the dual comparisons, such as W3-W5 and professional-trainee correlations, and two-way anova analysis (Matlab anovan()) to investigate the impact of two factors on impedance measures: expertise (professional/trainee/novice) and direction (v n /x/y/z) via comparing rh, m h , d h , and k h measures of the professional, trainee, and novice participants. Then, post-hoc analysis was performed with Tukey's test (multcompare()) to find the groups that significantly differ from each other with respect to a factor that shows a significant impact. In all the statistical tests throughout the paper, p = 0.05 was used as the threshold (maximum) for the statistical significance.   Table III where an overall improvement is indicated in all aspects with the novice subjects while proceeding throughout the training program. The more illustrative performance comparisons between the groups is shown in Fig. 11 where gradual performance improvement is taking place among the novice subjects and expertise level of the surgeons became apparent. It must be noted that, we do not claim that novice participants have approximately reached the proficiency level of the surgeons after the training program, yet one can state that they simply improved their laparoscopic skills only in a pre-planned suturing practice.

A. Statistical Analyses Among the Novice Subjects
Initially, we analysed statistical significance difference between the impedance measurements of the novice subjects by focusing on the data based on the Week 3 and Week 5 experiments. We excluded the Week 1 measurements because, as it was the first time they used such a system, it may well be argued that in Week 1 they had mainly focused on how to operate the tools rather then to the task itself.
The obtained results are given in Table IV where (−) indicates when there is no statistically significant difference between the weeks and W 5 > W 3 (W 5 < W 3 ) indicates when there exists statistically significant difference such that mean of the Week 5 measurements is higher (smaller) than the Week 3 measurements.

1) Rate-Hardness Values:
Rate-hardness of the left hand in Week 3 was found to be significantly smaller than the Week 5 measurements in the x direction (p = 0.019338). Statistically significant difference is not observed with respect to the right hand.
2) Mass Values: Left hand mass estimations in Week 3 were found to be significantly smaller than the Week 5 estimations in x (p = 1.38 × 10 −4 ) and in z (p = 0.0103) directions, yet it was vice versa in y (p = 2.06 × 10 −4 ) direction. On the contrary, the right hand's Week 3 was found to be significantly smaller than the Week 5 estimations in y (p = 0.0015) direction. 3

B. Statistical Analyses Between The Professional and Trainee Surgeons
Statistical significance difference analyses were carried out among the surgeons by comparing only impedance measurements of the professional and trainee surgeons. With this respect, it was observed that in all the significance difference results the measured mean hand-impedances of the professional surgeons were higher than the trainees' measurements (P > T ), see Table V.

1) Rate-Hardness Values:
In left hand, there exist statistically significant difference both in x (p = 6.4 × 10 −7 ) and z (p = 0.0009) directions; rate-hardness values of the professional surgeons' hands were higher than the trainees values.
2) Mass Values: The only statistically significant results were obtained within mass values of the professional and trainee surgeons is in x direction, which is parallel to the suturing line, both with left (p = 3.4 × 10 −7 ) and right (p = 0.0316) hands' measurements.
3) Damping Values: As in the mass measurements, the only statistically significant results were obtained within the damping values of the professional and trainee surgeons in the x direction both for left (p = 9.2 × 10 −8 ) and right (p = 0.0314) hands.

4) Stiffness Values:
The stiffness of the professional surgeons' left arm was found to be significantly higher than the trainees' only in the z direction (p = 0.0071).
C. Relativity between the Impedances of the Professional, Trainee Surgeons, and Novice Participants As a final effort, statistical significance analyses were carried out to compare the impedance measurements of the professional, trainee, and novice participants. Based on the overall performance calculations, we have used the Week 5 measurements within the forthcoming analyses to stand for impedances of the novice subjects. Also, for clarity the critical p values for the factors expertise and direction are stated as p e and p d , respectively.

1) Rate-Hardness Values:
For the rate-hardness measures, the two-way ANOVA found significant effect due to the factors expertise (p e ) and direction (p d ) both in left (p e = 1.7 × 10 −10 and p d = 4.5 × 10 −34 ) and right (p e = 0.0402 and p d = 1.7 × 10 −5 ) hands and there were significant interactions between these two factors within the both hands. The rate-hardness of the professional surgeons' left hand measure was found to be significantly higher than that of the trainee surgeons x (p = 4.8 × 10 −5 ) and z (p = 0.0109) directions. Similarly, it is also significantly higher than that of the novice subjects (p = 2.9 × 10 −7 ) in the x direction. We did not find significant difference between the professional, trainees, and novice participants in v n and y directions with the left hand measures. Regarding to the right hand measurements, the rate-hardness of the professional surgeons' was found to be significantly higher than that of the novice subjects only in v n (p = 0.0081) direction.
2) Mass Values: The two-way ANOVA analyses found significant effect due to the factors expertise and direction both in left (p e = 1.5 × 10 −6 and p d = 1.09 × 10 −22 ) and right (p e = 2.1 × 10 −5 and p d = 0.0434) hands within the mass measures, besides there were significant interactions between the two factors within the both hands. The mass of the professional surgeons' left hand measure was found to be significantly higher than that of the trainee surgeons (p = 6 × 10 −7 ) and Week 5 measurements (p = 5.8 × 10 −4 ) in x direction. But, with the right hand, Week 5 has the highest mass measurement in x direction, namely the right hand mass measure in Week 5 was found to be significantly higher than that of the professional (p = 2.5 × 10 −4 ) and the trainee (p = 2.7 × 10 −7 ) surgeons. In y direction, there exists statistical significance difference with left hand mass measures, such that the professional surgeons' measure was found to be significantly higher than that of the novice subjects' (p = 0.0181), yet we did not found any significance difference with right hand in the same direction. Similar to that, no significant difference observed in v n and z directions for the both hands' mass measurements.
3) Damping Values: With the damping measurements, the two-way ANOVA analyses found significant effect due to the factors expertise and direction both in left (p e = 5.5 × 10 −14 and p d = 1.06 × 10 −111 ) and right (p e = 0.0355 and p d = 1.4 × 10 −42 ) hands, besides there were significant interactions between these two factors within the both hands. The damping of the trainee surgeons' left hand was found to be significantly higher than that of the novice subjects in v n direction (p = 0.0136), yet no statistical difference observed with the right hand measures in this direction.
As in the mass analyses, the damping of the professional surgeons' left hand measure was found to be significantly higher than that of the trainee surgeons (p = 3.4 × 10 −7 ) and Week 5 measurements (p = 2.2 × 10 −7 ) in the x direction. But, with the right hand, Week 5 has the highest mass measurement in the x direction, namely the right hand damping measure in Week 5 was found to be significantly higher than that of the professional (p = 8 × 10 −7 ) and the trainee (p = 2.3 × 10 −7 ) surgeons. In the y direction, there exists statistical significance difference with left hand mass measures, such that the professional surgeons' measure was found to be significantly higher than that of the novice subject (p = 0.0167) and the trainees' measure was found to be significantly higher than that of the novice subject (p = 0.0018) as well. Yet, we did not observe any significance difference with right hand in this direction. Similar to that, no significant difference observed in z directions for the both hands' damping measurements.

4) Stiffness Values:
The two-way ANOVA analyses found significant effect due to the factor expertise in left hand (p e = 3.24 × 10 −4 ) and with the right hand due to factor direction (p d = 0.0101) within the stiffness measures, besides there was significant interactions between these two factors only within the right hand.
The Table VI provides the results obtained from the two-way Anova analyses of the impedance measurements grouped with respect to the expertise level of the participants.
V. DISCUSSION In our measurements, right hand-impedance was consistently found to be higher than the left hand's; Welch's t-test was carried out to compare impedance measurements of the left and right hands (p ≤ 8.2 × 10 −19 ∀ cases) with the data consisting of measurements in Week 5 and experiments with professional and trainee surgeons, see Table II as well. We hypothesize a link between that and difference of the left and right hands' instruments that typically require different grasps and therefore have diverse finger and hand postures. Previous research has established that hand-impedance can vary depending on stiffness of the hand grip [24]; this aspect, for instance, is frequently revisited in stability analyses of the bilateral teleoperation systems (see, e.g., [52]).
More number of statistically significant difference in the left hand-impedance parameters was observed compared to the right hand, across the experts and novices within the different levels of training, see Tables IV, V, and VI. We hypothesise that, this difference might be because the left hand instrument has been used more actively during needle placement and requires better control through both wrist and finger movements. Statistically significant and consistent difference was observed particularly in the x direction while comparing the expertise levels of all the participant groups. This is the direction, where the subjects were facing, parallel to the suturing line and more or less along the instrument shafts in the Cartesian plane. This consistence difference across the skill levels might be an indicator that the task involves more movements on this axis compared to the other directions.
Among the mass, damping, and stiffness parameters, we found statistically significant differences with mass and damping. However, no statistically significant difference in the stiffness was observed except for a single instant (Table V, left hand, z direction). Stiffness is related to the active use of muscles and muscle strength. We hypothesise that the experts do not maintain skilled performance due to their muscle strength nor because they activate their muscles more than the others but mostly because they orient their body-arm-hands better than the novice subjects, by adapting a more ergonomic posture, as proposed in [53].
The previous study, [54], indicates that age (< 70) and gender (when eliminating the effect of the bodyweight) have little or no effect on the mechanical impedance characteristics of the human elbow joint. In this initial study, we also neglected the effects of the demographic and anthropometric factors on the impedance of the human upper limps, see [55], for instance, the effect of these factors on the orientational impedances. A future research direction is to investigate the null hypothesis of "gender and age does not affect the hand impedances during laparoscopy".

VI. CONCLUSION
To the best of authors knowledge, the present study has demonstrated, for the first time, hand-impedance measurements of the surgeons and novice subjects in the laparoscopic suturing practice. Also, we demonstrate that the measurements are effective to capture skill related differences across professional laparoscopy surgeons and novice subjects. One can consider that the presented technique and the identified values can be useful for the following purposes in future research: i) to inform laparoscopy training practices in order to optimally orient the arms to maintain optimal handimpedance, ii) to be used in co-manipulated robotic trainers in order to gradually teach to the trainees the optimal hand-impedance, iii) to provide a biologically based method of assessment of laparoscopy skills with hand-impedance measurements, and iv) to be used in the stability analysis of the coupled robot-human system in a co-manipulated robotic trainer/assistant application. Current results lead us to the hypotheses, hence, further effort is required to validate aforementioned observations and correlations.