A Novel Computerized Surgeon Machine Interface for Robot-Assisted Laser Phonomicrosurgery

Transcription

1 The Laryngoscope VC 2014 The American Laryngological, Rhinological and Otological Society, Inc. A Novel Computerized Surgeon Machine Interface for Robot-Assisted Laser Phonomicrosurgery Leonardo S. Mattos, PhD; Nikhil Deshpande, PhD; Giacinto Barresi, MS; Luca Guastini, MD; Giorgio Peretti, MD Objectives/Hypothesis: To introduce a novel computerized surgical system for improved usability, intuitiveness, accuracy, and controllability in robot-assisted laser phonomicrosurgery. Study Design: Pilot technology assessment. Methods: The novel system was developed involving a newly designed motorized laser micromanipulator, a touchscreen display, and a graphics stylus. The system allows the control of a CO 2 laser through interaction between the stylus and the live video of the surgical area. This empowers the stylus with the ability to have actual effect on the surgical site. Surgical enhancements afforded by this system were established through a pilot technology assessment using randomized trials comparing its performance with a state-of-the-art laser microsurgery system. Resident surgeons and medical students were chosen as subjects in performing sets of trajectory-following exercises. Image processing based techniques were used for an objective performance assessment. A System Usability Scale based questionnaire was used for the qualitative assessment. Results: The computerized interface demonstrated superiority in usability, accuracy, and controllability over the stateof-the-art system. Significant ease of use and learning experienced by the subjects were demonstrated by the usability score assigned to the two compared interfaces: computerized interface % versus state-of-the-art %. The objective analysis showed a significant enhancement in accuracy and controllability: computerized interface % versus state-ofthe-art %. Conclusions: The novel system significantly enhances the accuracy, usability, and controllability in laser phonomicrosurgery. The design provides an opportunity to improve the ergonomics and safety of current surgical setups. Key Words: Robotics, CO 2 laser, microlaryngeal surgery, surgeon interface. Level of Evidence: N/A Laryngoscope, 124: , 2014 INTRODUCTION Lasers increasingly occupy a dominant position in the treatment of pathological conditions in the larynx. They are often used as a precision tool to perform delicate ablation or cutting procedures. Figure 1 provides an overview of the surgeon interfaces in laser-based microsurgeries. One such example is the use of CO 2 lasers in laser phonomicrosurgery (LP), which involve a suite of complex otolaryngological surgical techniques for the treatment of minute abnormalities in the voice box. 1 The From the Department of Advanced Robotics, Istituto Italiano di Tecnologia (L.S.M., N.D., G.B.); and Department of Otorhinolaryngology, University of Genoa (L.G., G.P.), Genoa, Italy Editor s Note: This Manuscript was accepted for publication December 12, The research was carried out at the Italian Institute of Technology in Genoa. The experimental trials were conducted at the Otorhinolaryngology Department of San Martino Hospital in Genoa. This research has received funding from the European Union Seventh Framework Program FP7/ , Challenge 2 Cognitive Systems, Interaction, Robotics, under grant agreement lralp No The authors have no other funding, financial relationships, or conflicts of interest to disclose. Send correspondence to Leonardo S. Mattos, Department of Advanced Robotics, Istituto Italiano di Tecnologia, Via Morego 30, 16163, Genova, Italy. leonardo.demattos@iit.it DOI: /lary CO 2 surgical laser, coupled with a surgical microscope, is one of the main tools in LP. 2 Here, a mechanical micromanipulator is used to manually aim the laser beam from a typical operating distance of 400 mm. This requires the surgeon to have high psychomotor skills to overcome challenges exacerbated by poor operating ergonomics, suboptimal visualization, and difficult surgical site access. 3 Recent improvements to the traditional surgical setup include the addition of motorized laser scanning mirrors able to execute preprogrammed patterns. This functionality enables higher quality incisions and ablations in comparison with results achieved using a manually guided laser beam. 4,5 At least three devices of this type are commercially available: the Lumenis (Tel Aviv, Israel) AcuBlade, the KLS Martin (Jacksonville, FL) SoftScan, and the DEKA (Manchester, NH) HiScan. However, these systems do not actually enhance the surgical user interface, as they still require the surgeon to operate while looking through the microscope and position the scan patterns on the desired targets using the same traditional micromanipulators. Figure 2 shows the state-ofthe-art surgical setup in LP, using the AcuBlade system. It is recognized that interfaces (and human factors) play a major role in the success and quality of laser surgery. 6 This article describes the research toward improving the precision, controllability, safety, and ergonomics 1887

2 Fig. 1. User interfaces for laser microsurgery: (a) Diomax handpiece, a handheld optical scalpel commercialized by KLS Martin; (b) UniMax 2000 by Reliant Laser (Agawam, MA), a passive laser micromanipulator for surgical microscopes; (c) da Vinci System by Intuitive Surgical (Sunnyvale, CA), which has been experimentally outfitted with an optical fiber for laser surgeries; (d) the KU Leuven (Leuven, Belgium) writing interface for robot-aided laser surgery based on a graphics tablet; and (e) the new Robot-Assisted Laser Phonomicrosurgery interface based on a tablet personal computer. [Color figure can be viewed in the online issue, which is available at for laser microsurgery. This is done with a redesign of the laser micromanipulator and the implementation of a new computerized surgeon machine interface, the Robot-Assisted Laser Phonomicrosurgery (RALP) system. The concept allows surgeons to perform operations using a graphics stylus over live video of the surgical Fig. 2. The state-of-the-art surgical setup in laser phonomicrosurgery. [Color figure can be viewed in the online issue, which is available at

3 Fig. 3. The Robot-Assisted Laser Phonomicrosurgery concept. The graphics pen controls in real time the aiming and activation of the surgical laser by touching the live microscope video shown on the interactive display. [Color figure can be viewed in the online issue, which is available at site. This system eliminates hand eye coordination problems and makes full use of surgeons skills without requiring operational training on the equipment. In addition, it greatly improves laser aiming precision and consistency with a custom-designed motorized laser micromanipulator. This provides for a 4-lm positioning accuracy and a mm range at the typical 400-mm operating distance, allowing the physician to realize long cuts with unprecedented levels of precision and uniformity. This article introduces the new RALP system and presents a pilot technology assessment through its comparison with a state-of-the-art laser microsurgery system, the Lumenis AcuBlade. A combination of objective (quantitative performance) and subjective (self-reported) assessment methods is used with randomized trials for the analysis. MATERIALS AND METHODS The RALP Concept The RALP concept 7 is shown in Figure 3. Here, the stylus controls the aiming and activation of the surgical laser in real time. This is done through interaction with the touch-screen monitor, where live video of the surgical site, captured with the microscope camera, is displayed. A Zeiss (Oberkochen, Germany) microscope (OPMI Sensera /S7 with xenon lamp) is used in the setup with a Karl Storz (Tuttlingen, Germany) HD camera attached. Software processing and control allow precise positioning of the laser spot with the stylus tip when it touches the screen. Visual feedback to the surgeon is augmented with cues and safety fixtures, as seen in Figure 3. Furthermore, the system is configured to keep the surgeon in full control of the surgical operations, so CO 2 laser activation requires an enabling signal from a footswitch. This system eliminates hand eye coordination problems and augments surgeons fine manual skills through gesture scaling and magnification. Safety and controllability are also improved, in this case via assistive teleoperation and automation techniques. The new system allows for the definition of virtual safety regions wherein the laser is automatically switched off. Finally, it also offers the option of programming scan patterns for the laser beam, which is also done using the stylus over the live video display. These are automatically executed by the control software at user-definable speeds, ensuring precise repeatability and high consistency on operations that require a number of laser passes over the same area. Motorized Laser Micromanipulator The motorized laser micromanipulator was designed to be attached to the state-of-the-art LP equipment and is installed on its laser entry port. 8 Figure 4 shows the working principle. The final laser micromanipulator device is fully computercontrolled and able to realize accurate scanning motions at frequencies up to 200 Hz. The scanning hardware is mounted into a custom-designed case (Fig. 4) that guarantees the necessary optical alignment. Validation trials with this motorized micromanipulator demonstrated an average trajectory-following error of only mm, a 70% reduction compared to the traditional manual micromanipulator. Experiment Design for Pilot Technology Assessment To validate the novel surgeon machine interface, the laser manipulation in LP with RALP was compared with the state-ofthe-art digital AcuBlade laser micromanipulator in a pilot technology assessment. The experiments in this research were designed following well-established norms for ergonomic evaluation and device usability analysis. The experimental tasks included sets of trajectory-following exercises, that is, surgical maneuvers to follow preset random shapes, including straight lines, C-curves, and S-curves, which require dexterity and good laser controllability. They were stamped on small plaster blocks with randomized sequences of shapes and shape orientations. Figure 5 shows the experimental setup. The trials were performed at San Martino Hospital (Genoa, Italy). At this stage, the assessment did not consider efficacy or morbidity using the two interfaces and did not use tissue samples (ex vivo or in vitro) for the trials. The trials do not involve real surgical tasks, and considered only the initial learning steps with emulated maneuvers. These aspects are part of future work in our research. 1889

4 Fig. 4. The new motorized laser micromanipulator (Fast Steering Mirror [FSM]) system: (a) working principle; (b) photograph of the final FSM device. [Color figure can be viewed in the online issue, which is available at Subjects and Groups Resident surgeons and medical students were chosen as subjects for the assessment. After classifying the subjects based on gender and educational level, they were divided into two groups; the AcuBlade group was composed of 16 subjects (average age years; 6 male, 10 female); the RALP group was Fig. 5. Experimental setup, video snapshot from a trial, and details of the precision targets stamped on a plaster block. RALP 5 Robot- Assisted Laser Phonomicrosurgery. [Color figure can be viewed in the online issue, which is available at

5 TABLE I. SUS Questionnaire Items. 9 TABLE II. Imaging-Based Metrics and Unified Rating. 11 Item Question Metric Description 1 I think that I would like to use this system frequently. 2 I found the system unnecessarily complex. 3 I felt very confident using the system. 4 I found the various functions in this system were well integrated. 5 I thought the system was easy to use. 6 I thought there was too much inconsistency in this system. 7 I found the system very cumbersome to use. 8 I would imagine that most people would learn to use this system very quickly. 9 I think that I would need the support of a technical person to be able to use this system. 10 I needed to learn a lot of things before I could get going with this system. SUS 5 System Usability Scale. composed of 16 subjects (average age years; 6 male, 10 female). Subjects were randomly chosen to be part of one group or the other. The groups were balanced for gender, and as far as possible for educational level. The two groups were made to obtain data about the effects of the two different interfaces on the subjects. There was no subjective bias in the choice of the subjects. The only criteria for selection were that the subjects should have the background and motivation to learn to use the surgeon machine interfaces (medical students) as well as good knowledge of surgical procedures (resident surgeons). By only including subjects without experience on the specific devices tested, we avoided biases in their performances using a particular system over another. All the subjects being nonexperts allowed us to understand the effects that the two interfaces had on the usability, performance, and mental workload of the users. This allowed an unbiased comparison of the interfaces. In the AcuBlade group, the experimental setup with the surgical equipment had the same configuration as in the operating room (Fig. 5, left). The subjects used their nondominant hand for laser manipulation, as per current medical practice. In the RALP condition, the subjects employed the novel user interface, as shown in Figure 5 (top, right). The subjects performed the trials in two sessions of two target blocks each, with a break of at least 10 minutes between the two. A within-group experimental design was not considered at this stage; the goal was to focus the study on the comparison between the two interfaces. The implemented experimental design avoided the effect of the prior experience of the subjects with the other interface. AR Ratio of the area covered by the laser-traced and desired shapes. PR Ratio of the boundary lengths of the laser-traced and desired shapes. ARM Ratio of the aspect ratios of the laser-traced and desired shapes. OM Absolute difference in orientations of laser-traced and desired shapes. RMSE Root mean square error method for distance between the laser-traced and desired shapes. Error max Maximum value of the distance between the lasertraced and desired shapes. f f AR 1 f PR 1 f ARM 1 f OM 1 f RMSE 1 f Errormax. Execution Time taken in laser-tracing the given shape. time AR 5 area ratio; ARM 5 aspect ratio measure; Error max 5 maximum path error; f 5 unified rating; OM 5 orientation measure; PR 5 perimeter ratio; RMSE 5 path following error by root mean square error. Execution time. The correlation of the subjective selfevaluation and quantitative measurement of total time spent in performing the trials can indicate the effect the interfaces have on the mental workload of the subjects. For this purpose, a question regarding the perceived time spent performing the trial was included in the questionnaire. The actual value of the time spent was obtained from the objective measures. RESULTS Statistical tests were used for the analysis of significance in the data. Figure 6 shows representative trials. Subjective Analysis The detailed list of the comparisons of SUS scores is presented in Table III. The average SUS global scores demonstrated that the subjects assigned a much greater overall usability for the RALP at 83.96% (standard Analysis Methodology Subjective measures. User-centered interface evaluations allow assessment of usability, ergonomics, and mental workload. For the subjective evaluations, the usability analysis questionnaire was used, 9 based on the System Usability Scale (SUS). 10 SUS provided a global view of usability through the assessment of aspects such as ease of use, ease of learning, et cetera, 9 as listed in Table I. Objective measures. To establish a quantitative basis for the evaluation of the two conditions, imaging-based metrics were used, as introduced. 11 The metrics are listed in Table II. A weighted combination of the metrics, a unified rating, was used to classify the performance of each trial by each subject. 11 Fig. 6. Representative sample from the trials. Top: The desired and laser-traced shapes for the AcuBlade condition. Bottom: The desired and laser-traced shapes for the Robot-Assisted Laser Phonomicrosurgery (RALP) condition. [Color figure can be viewed in the online issue, which is available at

6 TABLE III. Comparison of Average SUS Scores Over the Complete Set of Trials, 16 Subjects per Interface, 48 Trials per Subject. Subscale RALP Condition, Mean, % AcuBlade Condition, Mean, % Improvement for RALP Over AcuBlade, % Global score Subscale Subscale Subscale Subscale Subscale Subscale Subscale Subscale Subscale Subscale The 10 subscales correspond to the 10 questions that form the SUS. RALP 5 Robot-Assisted Laser Phonomicrosurgery; SUS 5 System Usability Scale. deviation [SD] %) over the AcuBlade at 68.02% (SD %). Student t test shows P 5.004, affirming the statistical significance of the improvement. The subscales in the SUS further illustrate the different aspects of usability for the interfaces. Because the subscale scores are human-generated, they do not allow the assumptions of normality and homoscedasticity, precluding the use of Student t test. The following observations are derived from an explorative comparison of the subscales: 1. The users of RALP feel more confident (subscale 3) than the users of AcuBlade during the tasks. 2. The RALP interface is easier to use (subscale 5) than the AcuBlade interface. 3. The RALP interface is easier to learn (subscale 8) than the AcuBlade interface. TABLE IV. Comparison of Average Values of the Metrics and the Unified Rating Over the Complete Set of Trials, 16 Subjects per Interface, 48 Trials per Subject. Metric RALP Condition AcuBlade Condition Raw Rating Raw Rating t for Rating P for Rating AR PR* e-6 ARM* e-4 OM* e-4 RMSE* e-6 Error max * e-9 Unified rating, %* e-8 *Statistically significant subscales. AR 5 area ratio; ARM 5 aspect ratio measure; Error max 5 maximum path error; OM 5 orientation measure; PR 5 perimeter ratio; RALP 5 Robot- Assisted Laser Phonomicrosurgery; RMSE 5 path following error by root mean square error. 4. The users of RALP would require less support by an expert (subscale 9) than the users of AcuBlade. 5. The users of RALP would require learning fewer processes (subscale 10) than the users of AcuBlade. Conversely, the RALP system also performs slightly worse than the AcuBlade with respect to preferred frequency of usage (subscale 1) and inconsistency in the system operations (subscale 10). This is taken as important feedback toward simplifying the user interface presented to the surgeons during surgery. Objective Analysis Each individual laser-traced shape contributes to the quantitative analysis, giving 768 data points for each interface. Table IV summarizes the raw values of TABLE V. Observations Derived From Values of Specific Metrics for the Two Conditions. RALP Condition AcuBlade Condition Metric Mean SD Mean SD Observations PR The laser-traced shapes under the RALP condition conform to the length and thickness of the desired shapes significantly better than in the AcuBlade condition. ARM The RALP condition allows better shape conformance and positioning with the laser than the AcuBlade condition. OM, degrees The laser-traced shapes under the RALP condition are better aligned with the desired shapes than in the AcuBlade condition. RMSE, mm The RALP condition results in more precise laser control than the AcuBlade condition. Error max, mm The RALP condition permits safer control of the laser than the AcuBlade condition. It minimizes deviations from the desired trajectories. ARM 5 aspect ratio measure; Error max 5 maximum path error; OM 5 orientation measure; PR 5 perimeter ratio; RALP 5 Robot-Assisted Laser Phonomicrosurgery; RMSE 5 path following error by root mean square error; SD 5 standard deviation. 1892

7 the individual metrics and their contribution to the unified rating scores. Student t test analysis was performed on the unified rating. The RALP condition presents a significantly better performance at 90.02% (SD ) than the AcuBlade condition at 75.59% (SD ; P 5 9.3e-8). An analysis of the raw values of the individual metrics presents a clearer difference in the performance impact of the two interfaces, as summarized in Table V. Welch t test was used for the individual metrics due to the lack of homoscedasticity in the variances of their distributions. 12 The metrics in Table V are significant because they are directly related to the safety and efficiency of the interfaces. A high value for the orientation measure indicates that the majority of the laser-traced shape is not aligned with the desired shape. A high value for the root mean square error indicates imprecise laser control. Similar inferences are drawn from the perimeter and aspect ratio metrics. In the case of real surgery, this is highly undesired, impacting the safety and quality of procedures. On these parameters, the RALP system consistently shows closer to ideal values, much more so than the AcuBlade interface. Execution Time Analysis To understand the effect of the two interfaces on the perceived and actual times for trial completion, the Mann-Whitney U test was used, due to the lack of normality in the distributions. In subjective evaluations, there is a significant difference for the trial times in the two conditions (P 5 2e-4). The subjects in the AcuBlade condition estimated a longer time taken to complete the trials (mean seconds, SD ) than the subjects in the RALP condition (mean seconds, SD ). The quantitative analysis is contrary to the subjective evaluation. It shows a significant difference (P 5.002), but the RALP condition (mean seconds, SD ) takes a longer time to complete than the AcuBlade interface (mean seconds, SD ). This is an interesting result, as it allows inferring the level of mental workload necessitated in each condition. The inverse correlation between the perceived and actual times indicates the subjects in the RALP condition perceive themselves to have spent less time performing the trials than the subjects in the AcuBlade condition. The RALP condition seems to induce less mental workload than the AcuBlade condition. This is advantageous for the acceptance and usability of the new RALP system, indicating it can also help reduce mental demands imposed on the physician. CONCLUSION This article presented a novel medical robotic system, the RALP system, created to provide better precision, controllability, safety, and ergonomics for LP procedures. These objectives were achieved with the implementation of the RALP concept and creation of a custom motorized laser micromanipulator. The novel interface was compared Fig. 7. The unified rating and the System Usability Scale (SUS) global scores for the AcuBlade and the Robot-Assisted Laser Phonomicrosurgery (RALP) conditions. The means and the standard deviations are shown as percentages. [Color figure can be viewed in the online issue, which is available at www. laryngoscope.com.] to the current, state-of-the-art surgeon interface in LP. The combination of the subjective SUS-based evaluation and the quantitative metrics-based evaluation provide for a clear classification of the two examined surgical interfaces from the usability and performance perspectives. The results are summarized as follows: 1. In the RALP system, delicate maneuvers can be accurately performed from a computer monitor using a graphics stylus directly over the live video captured from the surgical microscope. This was shown to eliminate typical hand eye coordination problems and make better use of surgeons manual dexterity. 2. The subjective evaluation shows a superior overall usability for the RALP system over the AcuBlade interface. The global SUS score of 83.96% for the RALP system is significantly higher, by about 23%, than that for the AcuBlade interface at 68.02%. The advantages of the RALP interface are demonstrated through five different subscales of the SUS scale, which indicate that it is an easy to use and easy to learn interface. 3. The objective evaluation points to the clear advantage of the RALP interface over the AcuBlade interface. The analysis showed the significant superiority of RALP in terms of the efficiency and safety aspects of its performance. With a unified rating of 90.02%, it compared advantageously, about 19% better, against the AcuBlade interface with a unified rating of 75.59%. Figure 7 captures the values for the quantitative and qualitative analysis in a histogram plot. It is evident from the plot that the RALP system shows an overall higher performance and usability rating than the Acu- Blade condition. In the extension of this research, further studies are planned to better understand the usability of surgical interfaces through an integrated framework for evaluation. The technology advances presented here are now in the process of being evaluated jointly by surgeons and engineers using ex vivo pig larynxes. This will provide the groundwork for further evaluation with animal trials and future clinical application. Continuing interactions with the surgeons and other experts in the ergonomics 1893

8 field shall further help adapt the system architecture for the operating room. BIBLIOGRAPHY 1. Rosen J, Solazzo M, Hannaford B, Sinanan MN. Task decomposition of laparoscopic surgery for objective evaluation of surgical residents learning curve using hidden Markov model. Comput Aided Surg 2002;7: Solares CA, Strome M. Transoral robot-assisted CO2 laser supraglottic laryngectomy: experimental and clinical data. Laryngoscope 2007;117: Dagnino G, Mattos LS, Caldwell DG. New Software Tools for Enhanced Precision in Robot-Assisted Laser Phonomicrosurgery. In Proceedings of 34th International Conference of IEEE Engineering in Medicine and Biology Society. 2012: , IEEE, San Diego, CA, USA. 4. Remacle M, Hassan F, Cohen D, Lawson G, Delos M. New computerguided scanner for improving CO2 laser-assisted microincision. Eur Arch Otorhinolaryngol 2005;262: Remacle M, Lawson G, Nollevaux M, Delos M. Current state of scanning micromanipulator applications with the carbon dioxide laser. Ann Otol Rhinol Laryngol 2008;117: Wong Y-T, Finley CC, Giallo JF Jr, Buckmire RA. Novel CO2 laser robotic controller outperforms experienced laser operators in tasks of accuracy and performance repeatability. Laryngoscope 2011;121: Mattos LS, Dagnino G, Becattini G, Dellepiane M, Caldwell DG. A virtual scalpel system for computer-assisted laser microsurgery. In: Proceedings of IEEE/RSJ International Conference on Intelligent Robots and Systems. 2011: , IEEE, San Francisco, CA, USA. 8. Mattos LS, Dellepiane M, Caldwell DG. Next generation micromanipulator for computer-assisted laser phonomicrosurgery. In: Proceedings of 33rd International Conference of IEEE Engineering in Medicine and Biology Society. 2011: , IEEE, Boston, MA, USA. 9. Barresi G, Deshpande N, Mattos LS, et al. Comparative usability and performance evaluation of surgeon interfaces in laser phonomicrosurgery. In: Proceedings of IEEE/RSJ International Conference on Intelligent Robots and Systems. 2013: , IEEE, Tokyo, Japan. 10. Brooke J. SUS: a quick and dirty usability scale. In: Jordan PW, Thomas B, Weerdmeester BA, McClelland AL, eds. Usability Evaluation in Industry. 1996: , Taylor & Francis, London, UK. 11. Deshpande N, Mattos LS, Barresi G, et al. Imaging based metrics for performance assessment in laser phonomicrosurgery. In: Proceedings of International Conference on Robotics and Automation. 2013: , IEEE, Karlsruhe, Germany. 12. Welch BL. The generalization of Student s problem when several different population variances are involved. Biometrika 1947;34: