- Email: [email protected]
Patient-specific Virtual Simulation—A State of the Art Approach to Teach Renal Tumor Localization
Accepted Manuscript
Patient-specific virtual simulation - a state of the art approach to teach renal tumor localization Arun Rai , Jason M. Scovell , Ang Xu , Adithya Balasubramanian , Ryan Siller , Taylor Kohn , Young Moon , Naveen Yadav , Richard E. Link PII: DOI: Reference:
S0090-4295(18)30540-5 10.1016/j.urology.2018.04.043 URL 21063
To appear in:
Urology
Received date: Revised date: Accepted date:
7 January 2018 19 April 2018 27 April 2018
Please cite this article as: Arun Rai , Jason M. Scovell , Ang Xu , Adithya Balasubramanian , Ryan Siller , Taylor Kohn , Young Moon , Naveen Yadav , Richard E. Link , Patient-specific virtual simulation - a state of the art approach to teach renal tumor localization, Urology (2018), doi: 10.1016/j.urology.2018.04.043
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
Patient-specific virtual simulation - a state of the art approach to teach renal tumor localization Arun Rai1* and Jason M. Scovell1,2*, Ang Xu1, Adithya Balasubramanian1, Ryan Siller1, Taylor Kohn1, Young Moon1, Naveen Yadav1, Richard E. Link1,3 1
Scott Department of Urology, 2Translational Biology and Molecular Medicine and 3Dan
USA. *AR and JMS contributed equally to this work
AN US
The authors declare no conflict of interest Abstract word count: 249 Manuscript word count: 2929 Corresponding author: Richard E. Link
CR IP T
L. Duncan Comprehensive Cancer Center, Baylor College of Medicine, Houston, TX,
PT
ED
M
Richard E. Link, MD, PhD Carlton-Smith Endowed Chair in Urologic Education Associate Professor of Urology Scott Department of Urology Baylor College of Medicine 7200 Cambridge, MC BCM380, A10.107 Houston, TX 77030, U.S.A. Tel: (713) 798-7670 Fax: (713) 798-5553 Email: [email protected]
Key Words: education, algorithm, 3d printing, robotic surgery, partial nephrectomy
CE
Running title: Patient-specific virtual simulation - a state of the art approach to teach renal tumor localization
AC
Source of Funding: Research reported in this publication was supported by the National Institute of General Medical Sciences of the National Institutes of Health under Award Number T32GM088129 as well as the loan of the dV-Trainer simulator platform from Mimics Technologies, Inc. (Seattle, WA). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health or Mimics Technologies.
ACCEPTED MANUSCRIPT
ABSTRACT OBJECTIVES: To test a novel visuospatial testing platform improve trainee ability to convert two-dimensional (2D) to three-dimensional (3D) space.
CR IP T
METHDOS: Medical students were recruited from Baylor College of Medicine and McGovern Medical School (Houston, TX). We 3D reconstructed three partial
nephrectomy cases using a novel, rapid and highly accurate edge-detection algorithm. Patient-specific reconstructions were imported into the dV-Trainer (Mimics
AN US
Technologies, Seattle, WA) and used to generate custom 3D printed physical models. Tumor location was altered digitally to generate nine physical models for each case, one with the correct tumor location and eight with sham locations. Subjects were randomized 1:1 into the dV-Trainer (intervention) and No-dV-Trainer (control) groups.
M
Each subject completed the following steps: (1) visualization of computed-tomographic
ED
images, (2) visualization of the reconstructed kidney and tumor in the dV-Trainer (intervention group only), and (3) selection of the correct tumor location on the 3D
PT
printed models (primary outcome). Normalized distances from the correct tumor location
CE
were quantified and compared between groups. RESULTS: 100 subjects were randomized for this study. dV-Trainer use significantly
AC
improved subjects ability to localize tumor position (Tumor Localization Score: 0.24 vs. 0.38, p<0.001). However, subjects in the No-dV-Trainer group more accurately assigned R.E.N.A.L. CONCLUSION: Even brief exposure to interactive patient-specific renal tumor models improves a novice’s ability to localize tumor location. Virtual reality simulation prior to
ACCEPTED MANUSCRIPT
surgery could benefit trainees learning to localize renal masses for minimally invasive partial nephrectomy.
CR IP T
INTRODUCTION
One of the major technical hurdles a novice faces early in surgical training in developing the ability to train their visuospatial ability. Visuospatial ability refers to the capacity for mentally visualize and/or manipulate objects (4). Many surgical diagnoses are made by
AN US
standard two dimensional (2D) imaging such as Computed Tomography (CT) and
intervention requires the surgeon to translate these findings into a three dimensional (3D) mental image. This challenge is further amplified during minimally-invasive surgery
M
such as laparoscopic or robotic surgery where anatomic evaluation is indirect (via camera) and there is a lack or complete absence of direct tactile feedback. Not
ED
surprisingly, a recent meta-analysis concluded that performance on visuospatial aptitude, and in particular dynamic three dimensional testing, correlates with
PT
laparoscopic skills.1
CE
To address the importance of visuospatial ability, we have developed a novel training platform that utilizes a dynamic 3D virtual simulation environment (dV-Trainer) as a
AC
training bridge between 2D and 3D space. This platform evaluates visuospatial recognition using 3D printed models developed from standard 2D imaging. We have chosen robotic-assisted laparoscopic partial nephrectomy (RALPN) as a surgical test case for our novel training platform to assess visuospatial understanding. RALPN is a challenging procedure to learn and visuospatial understanding of correct
ACCEPTED MANUSCRIPT
tumor position, orientation and relationship to vital renal structures is essential for successful tumor excision and renal repair.5, 6 Safe and successful RALPN requires a subset of specialized skills that can be challenging to learn: sophisticated tumor visualization within a compact three dimensional (3D) space, proper tissue handling
CR IP T
despite lack of haptic feedback, and execution of tumor resection under ischemiadependent time constraints.5
We hypothesized that manipulating a patient-specific tumor/kidney model within a 3D
AN US
virtual simulator (dV-Trainer) would improve a novice’s ability to localize tumor location on a physical model (visuospatial ability) as compared to interpreting standard 2 dimensional planar imaging alone.
M
SUBJECTS AND METHODS
ED
Subjects
After obtaining Institutional Review Board (IRB) approval, medical students (years 1 to
PT
4) were recruited as subjects from Baylor College of Medicine and University of Texas McGovern Medical School in Houston, TX during 2016. Model kidneys used for surgical
CE
simulation were generated from cross-sectional computed-tomographic (CT) imaging of three patients who had previously undergone RALPN with surgeon R.E.L. at our
AC
institution. These three models were selected based on their diverse tumor locations and differences in Nephrometry score.10 Image Reconstruction and Virtual Model Generation
ACCEPTED MANUSCRIPT
The three patient-specific 3D models were reconstructed based on the available preoperative CT imaging (Figure 1), and were converted into both a virtual reality model within the dV-Trainer environment (intervention) and a physical 3D printed plaster model (primary outcome). The 3D reconstruction was generated through a novel unpublished
CR IP T
―edge-detect‖ MATLAB algorithm (MathWorks; Natick, MA) that utilizes image
processing to detect the edges of the kidney. This is made possible by the sudden
gradation of contrast enhancement on nephrogenic phase of the CT between renal parenchyma and surrounding fat which is then capable of generating a series of ―edges‖
AN US
through each axial CT section. This algorithm allows for rapid reconstruction of kidney and tumor in less than 60 seconds (Supplemental Video 1). Upon generating a series of edges for the entirety of the kidney, edges are retraced with CV curves in a plugin for a
M
3D modeling program. Each image is a cross sectional curve in the program and are then logged into an accurate 3D model of the kidney. Tumors were identified and
ED
marked in a similar fashion. The resulting in silico 3D model was saved in .stl file format and exported for final processing, refinement and importation into the dv-Trainer system
PT
(Mimic Technologies Inc., Seattle WA). This process is the first, to our knowledge, to be
CE
used to efficiently generate high-fidelity 3D reconstructions in an automated fashion. Upon generation of the 3D virtual reality model by Mimic Technologies (Seattle, WA),
AC
proofs underwent final approval for face validity with the authors prior to study initiation. Orientation markers were included in the dV-Trainer operating field for cranial, caudal, anterior and posterior. Virtual reality models could be rotated about the hilum and manipulated in an anterior/posterior fashion. Transparency of the renal tissue could be toggled on the desktop unit adjacent to the dV-Trainer console. The transparency toggle
ACCEPTED MANUSCRIPT
was used to help identify the location of the renal tumor within the substance of the kidney parenchyma. 3D ColorJet Printed Model Generation
CR IP T
Similar to the virtual reality models, the 3D printed models were generated based on the same 3D reconstructions generated via the ―edge-detect‖ algorithm. These models were 3D printed using ColorJet technology (ProJet 260C, 3D Systems, Rock Hill, SC) with the parenchyma as red and the tumor as green to help subjects delineate the mass
AN US
from the kidney. All plaster models were digitally bivalved in the sagittal plane prior to printing in order to help delineate anterior/posterior as well as endophytic/exophytic orientation. For each of the three renal models, 1 ―true‖ patient-specific model was generated in which the renal mass was located in the orthotopic position as delineated
M
from the CT imaging. For each patient-specific model, 8 additional variants were
ED
generated wherein the renal mass was localized to other randomized locations in the parenchyma. False were placed throughout the kidney – some tumors were placed very
PT
close (<1 cm away) and others were placed at distant locations (opposite pole). Upon the generation of these models, all tumor locations were assigned a tumor localization
CE
score (TLS) on a scale from 0 (center of the correct renal mass, orthotopic position) to 1
AC
(center of the furthest renal mass as measured as a straight line (i.e. if the furthest mass was noted to be 5 cm, a model with a tumor located 2.5 cm away was assigned a score of 0.5). Normalization was performed to allow quantification and enable comparisons across models. Study Design
ACCEPTED MANUSCRIPT
Prior to study initiation, all participants were asked to complete a questionnaire which included questions about demographics, post-graduate plans (surgical, non-surgical, or undecided), as well as three visuospatial aptitude questions in order to assess baseline 3D spatial reasoning (Supplemental Figure 3). The study was carried out in multiple
CR IP T
small groups in the Baylor College of Medicine Simulation Lab in order to minimize wait time and ensure study fluidity given the large number of participants. Informed consent was obtained from each participant in accordance with our IRB protocol. A short
orientation presentation was provided to participants detailing the components of the
AN US
study, how to interpret CT imaging of the kidney and a review of how to calculate the R.E.N.A.L. Nephrometry score.10 After this point, all participants were taken to the dVTrainer and completed two orientation exercises (Camera Targeting and Pick and
M
Place) in order to familiarize participants with the dV-Trainer as well as to collect user metrics. At this time, participants were randomized 1:1 into two groups – ―No dV-
ED
Trainer‖ and ―dV-Trainer‖ based on an A:B pattern. Both groups began first with Model #1 and were allowed 5 minutes to review the CT imaging (Supplemental Figure 1).
PT
Upon review of the CT imaging, the ―dV-Trainer‖ group was then allowed to review and
CE
manipulate the renal model in the dV-Trainer virtual reality trainer for 5 minutes. In comparison, the ―no dV-Trainer‖ group was taken immediately after randomization to a
AC
workstation wherein the 9 plaster models were laid out (Supplemental Figure 2). Five minutes were allotted to the participants to select the model they felt best represented the renal tumor visualized on CT imaging. After completion of 5 minutes on the dVTrainer, the ―dV-Trainer‖ group was taken to the same workstation to select the most representative physical model. After each model, all participants were then asked to fill
ACCEPTED MANUSCRIPT
out a questionnaire regarding the tumor’s Nephrometry score as well as which physical model best represented the mass they had viewed on CT imaging (primary outcome) (Supplemental Figure 4). This process was repeated for models 2 and 3.
CR IP T
Statistical Analysis A power analysis was performed a priori to detect a medium effect size (Cohen d = 0.6) using a Student’s t-test (β = 0.20, α = 0.05) and determined that a total of 45 subjects were required per group for a total of 90 subjects. The primary outcome for this study
AN US
was the tumor localization score between subjects in the ―dV-Trainer‖ and ―No dV-
Trainer‖ groups. Comparisons of proportions were performed using the Chi-Square test and Fisher test where appropriate. We compared continuous variables using the Student’s t-test. Univariate analysis was performed using the Generalized Linear Model
M
with tumor localization score as the dependent variable. Statistical analysis was
ED
performed using R v.3.3 on RStudio v.1.0 and SPSS v22 (IBM, NY).
PT
RESULTS
Three models were generated for the study and graded by Nephrometry score by R.E.L.
CE
(Model 1: 1+2+3+a+1 = 7a, left sided; Model 2: 1+2+3+p+3 = 9p, left sided; Model 3: 1+1+3+p+2 = 7p, right sided). 100 medical students were recruited and participated in
AC
the study. Subjects were randomized to the intervention group (dV-Trainer; n = 50), or the control group (No dV-Trainer; n = 50). Between groups, there was no difference in subject age (dV: 23.5 vs. No-dV: 23.8, p = 0.53), MS year (p = 0.60), and desired future specialty (p = 0.60) (Table 1). We assessed spatial reasoning using a 3-question spatial aptitude test and two warm-up exercises on the dV-Trainer (targeting, pick and place).
ACCEPTED MANUSCRIPT
There was no baseline difference in 3D aptitude between groups (p = 0.23) or ability with the dV-Trainer (p = 0.89, p = 0.12). Subjects in the dV-Trainer group demonstrated improved tumor localization compared
CR IP T
to subjects that did not use the dV-Trainer (Tumor Localization Score: 0.24 vs. 0.38, p<0.001) (Figure 2, Table 2). There was no difference in tumor localization for Model 1 (0.44 vs 0.49, p=0.43), but significant differences were observed for Model 2 (0.17 vs. 0.31, p=0.01) and Model 3 (0.12 vs. 0.34, p=0.001).
AN US
Contrary to our observation that dV-Trainer use improved subject ability to localize tumor location, subjects that did not use the dV-Trainer more accurately assigned R.E.N.A.L. Nephrometry scores (56% vs. 63%, p = 0.013). Subjects in the No-dV-
Trainer group more accurately assigned the correct diameter (78% vs. 91%, p = 0.004),
M
if the mass was endophytic/exophytic (36% vs. 53%, p = 0.004), and if the mass was
ED
anterior or posterior (87% vs. 69%, p < 0.001). Subjects in the dV-Trainer group were better able to assign location of the mass relative to the polar lines (68% vs. 47%, p <
PT
0.001). There was no difference in capacity to determine nearness to the collecting
CE
system with or without dV-Trainer use (31% vs. 35%, p = 0.46). We performed a univariate analysis to determine if any additional factors may associate
AC
with ability to identify the correct physical tumor location (Supplemental Table 1). We found that only dV-Trainer use was associated with improved tumor localization (β = 0.137, 95% CI: 0.070 – 0.203, p < 0.001). Age, 3D aptitude score, dV-Trainer warm-up scores, ability to assign the correct R.E.N.A.L. Nephrometry score, MS year, and desired future specialty did not associate with tumor localizing ability.
ACCEPTED MANUSCRIPT
DISCUSSION Surgical simulation can be divided into two broad categories – mechanical and virtualreality based simulation. Mechanical simulators include simple box trainers wherein
CR IP T
tasks are completed via videoscopic guidance. In virtual reality simulators, tasks are completed in a computer-generated environment often with 3D visualization.6 These virtual reality simulators have been developed for both robotic and traditional
laparoscopic kidney surgery. However, the validity of these simulators for improving
AN US
surgical performance remains controversial.
We hypothesized that even a short exposure to an interactive 3D virtual reality simulator (dV-Trainer) containing a patient-specific model could enhance a novice’s ability to create an accurate mental picture of tumor location. We tested this hypothesis by
M
asking the subject to identify the correct tumor location from a panel of different physical
ED
models containing variations in tumor location. We chose to recruit medical students to test this hypothesis due to their relatively uniform novice status with respect to
PT
visualizing renal masses from 2D planar images. The pool of medical students also provided the large number necessary to adequately power this study. Here we
CE
demonstrate that novice trainees who utilize the dV-Trainer virtual environment in
AC
addition to standard 2D planar imaging more accurately localize tumors in 3D physical models as compared to trainees that have access to 2D planar images alone. The R.E.N.A.L. Nephrometry score was developed to stratify the complexity of renal tumors based on several measurements of tumor anatomy (R—radius, E—exophytic vs endophytic, N—nearness of tumor to collecting system, A—anterior vs posterior, L—
ACCEPTED MANUSCRIPT
location relative to polar lines).10 Several studies have validated the utility nephrometry scoring in predicting perioperative and renal functional outcomes.13 More recent studies have demonstrated that radius, endophytic location and nearness to the collecting system are specifically associated with significant changes in warm ischemia time and
CR IP T
overall operative time.
We anticipated that access to the dV-Trainer would also enhance a novice trainee’s ability to accurately assign R.E.N.A.L. nephrometry scores. However, this was not
AN US
uniformly true. Despite improving the novice’s ability to localize a tumor on a 3D
physical model, use of the dV-Trainer decreased the accuracy with which subjects could attribute tumor diameter, how endophytic/exophytic, and how anterior or posterior the mass was for nephrometry scoring. Use of the dV-Trainer only increased subject ability
M
to properly localize where the tumor fell within polar lines. Perhaps, the apparent discrepancy between ability to localize a tumor on a physical model and assign the
ED
correct proper nephrometry score underlies the inherent complexity of converting the
PT
objective description obtained from 2D planar imaging to the real anatomical orientation and position of the tumor and mass. Certain aspects of the nephrometry score,
CE
specifically the nearness to the collecting system, require careful review and measurement on planar images that may require more than 5 minutes of study. This is
AC
one potential explaination for some of the reported discrepancies between nephrometry scores from urologists and radiologists for the same tumor case.14 We suspect that nephrometry scoring is a more challenging and time consuming exercise for novices than simply developing an accurate mental image of tumor location sufficient for surgical performance.
ACCEPTED MANUSCRIPT
3D printing has become more accessible, which is driving development of patientspecific physical models as surgical training tools. Knoedler et al. demonstrated that 3D printed physical models can enhance medical students’ ability to correctly assign the R, N, and L components of the Nephrometry score compared to standard 2D planar
CR IP T
imaging alone.15, 16 This highlights the potential educational differences between
physical and digital models, at least for novice trainees. Several groups, including our own, have also reported development of patient-specific soft-tissue models that can be utilized for pre-surgical rehearsal, or as part of a patient-specific ex vivo training
AN US
program.8, 17, 18 Clearly, these exciting new technologies have great potential to enhance surgical education and further randomized trials are needed to define the value of each type of model.
M
This study has several important strengths and weaknesses that warrant discussion. Our conclusions are strengthened by the prospective double-bind randomized study
ED
design and by sufficient accrual to satisfy our a priori power analysis. Our study also
PT
employed three relatively novel technologies: renal and tumor modeling using an automated point-and-click algorithm, importation of patient-specific models into a 3D
CE
dV-Trainer simulator, and 3D printing of digitally altered model variations. Our results support the hypothesis that even brief use of a 3D simulator can enhance a novice’s
AC
ability to identify renal tumor location in the physical world. Developing new technologies and simulators can be a costly and time-consuming enterprise that requires solid proof of concept to justify investment. By performing this study in a relatively homogeneous population of subjects, using a randomized approach, we intended to provide the best possible pilot data to assess the potential
ACCEPTED MANUSCRIPT
benefit of simulation in teaching tumor localization. Our intent was to determine whether or or not VR simulation can improve a novice’s visuospatial understanding of patient anaotmy. The physics and engineering for patient specific VR surgery does not exist yet and hence this study was designed and completed to help demonstrate the utility of VR
CR IP T
in its current state.
Academic urologists who teach trainees and all practicing urologists have a vested interest in how future trainees are taught to perform robotic surgery. A very large
AN US
proportion of urologic surgeons participate in teaching urology trainees. Many senior urologic surgeons who have only recently adopted robotics for partial nephrectomy likely struggle with the same visualization challenges faced by trainees when they switch from an open to a robotic environment for this procedure. Minimally invasive
M
surgery provides new opportunities for innovation in designing training regimens. Many of these innovative new approaches, whether they involve surgical simulation, the use
ED
of video games to teach surgery, 3D printed anatomic models, laparoscopic trainer
PT
boxes, and video based surgical evaluation are still developing. It is important for the major stakeholders to continue to explore, debate, and refine these methods as these
CE
techniques will drive how future urologists are trained.
AC
There are several important limitations to this study, including the novice study population (medical students) and relatively low-complexity nature of our models. While medical students are an excellent choice for a population of RALPN novices, these conclusions may not be translatable to a more experienced trainee cohort. Although we provided a standardized tutorial on how to interpret renal CT images prior to this study, it is difficult to determine how effective this tutorial was in establishing an imaging
ACCEPTED MANUSCRIPT
interpretation baseline for medical students. We suspect that more advanced trainees (i.e. surgical interns) would still benefit from using the dV-Trainer to localize renal masses; however, this effect may not be as substantial as we observed in the medical student population. Our models were generated using a novel point-and-click algorithm
CR IP T
which enabled rapid generation of digital models. This approach reduces a significant time-dependent barrier required for patient-specific surgical simulation. Although these models are visually similar to models generated by the gold-standard manual
segmentation technique, we did not rigorously quantify their similarity here. This is an
AN US
intended area of future research. These models represented only the patient’s renal parenchyma and tumor. In the dV-Trainer virtual reality atmosphere we added standard (non-patient specific) vasculature and ureter. Although difficult to determine why a
M
difference was noted in models 2 and 3 but not in model 1, we speculate that this was due to the random placement of a ―false‖ tumor very close to the ―true‖ location. As a
ED
result, participants in both groups struggled to differentiate between these two locations. Generating higher fidelity models that include perinephric fat, the renal collecting
PT
system, and adjacent abdominal organs may be beneficial when translating this
CE
application to patient-specific pre-surgical training.
AC
CONCLUSION
Visualizing patient-specific kidneys and tumors within the dV-Trainer 3D surgical simulation environment, even for only 5 minutes, improves a novice’s ability to localize correct tumor location in the physical world. Visual processing of 2D imaging into 3D anatomical structures is a challenging step for many surgeons. Use of the dV-Trainer as
ACCEPTED MANUSCRIPT
a mechanism to test and improve this ability may provide a powerful new approach for novices to adapt this skill towards a broad array of surgical procedures. REFERENCES
8. 9. 10. 11. 12.
AC
13.
AN US
7.
M
6.
ED
5.
PT
2. 3. 4.
Kramp KH, van Det MJ, Hoff C, Veeger NJ, ten Cate Hoedemaker HO, Pierie JP. The predictive value of aptitude assessment in laparoscopic surgery: a meta-analysis. Med Educ. 2016;50:409-427. SEER. Cancer Stat Facts: Kidney and Renal Pelvis Cancer2016. ACS. Cancer Facts & Figures 20172017. Luciani LG, Cestari R, Tallarigo C. Incidental renal cell carcinoma-age and stage characterization and clinical implications: study of 1092 patients (1982-1997). Urology. 2000;56:58-62. Ahmadi H, Liu JJ. 3-D Imaging and Simulation for Nephron Sparing Surgical Training. Current urology reports. 2016;17:58. Albani JM, Lee DI. Virtual reality-assisted robotic surgery simulation. Journal of endourology. 2007;21:285-287. Patel HD, Mullins JK, Pierorazio PM, et al. Trends in renal surgery: robotic technology is associated with increased use of partial nephrectomy. The Journal of urology. 2013;189:1229-1235. von Rundstedt FC, Scovell JM, Agrawal S, Zaneveld J, Link RE. Utility of patient-specific silicone renal models for planning and rehearsal of complex tumour resections prior to robot-assisted laparoscopic partial nephrectomy. BJU international. 2016. von Rundstedt FC, Scovell JM, Agrawal S, Zaneveld J, Link RE. Utility of patient-specific silicone renal models for planning and rehearsal of complex tumour resections prior to robot-assisted laparoscopic partial nephrectomy. BJU international. 2017;119:598-604. Kutikov A, Uzzo RG. The R.E.N.A.L. nephrometry score: a comprehensive standardized system for quantitating renal tumor size, location and depth. The Journal of urology. 2009;182:844-853. Leow JJ, Heah NH, Chang SL, Chong YL, Png KS. Outcomes of Robotic versus Laparoscopic Partial Nephrectomy: an Updated Meta-Analysis of 4,919 Patients. The Journal of urology. 2016;196:1371-1377. Choi JE, You JH, Kim DK, Rha KH, Lee SH. Comparison of perioperative outcomes between robotic and laparoscopic partial nephrectomy: a systematic review and meta-analysis. European urology. 2015;67:891-901. Watts KL, Ghosh P, Stein S, Ghavamian R. Value of Nephrometry Score Constituents on Perioperative Outcomes and Split Renal Function in Patients Undergoing Minimally Invasive Partial Nephrectomy. Urology. 2017;99:112-117. Benadiba S, Verin AL, Pignot G, et al. Are urologists and radiologists equally effective in determining the RENAL Nephrometry score? Annals of surgical oncology. 2015;22:16181624. Silberstein JL, Maddox MM, Dorsey P, Feibus A, Thomas R, Lee BR. Physical models of renal malignancies using standard cross-sectional imaging and 3-dimensional printers: a pilot study. Urology. 2014;84:268-272.
CE
1.
CR IP T
REFERENCES
14. 15.
ACCEPTED MANUSCRIPT
16. 17.
AC
CE
PT
ED
M
AN US
CR IP T
18.
Knoedler M, Feibus AH, Lange A, et al. Individualized Physical 3-dimensional Kidney Tumor Models Constructed From 3-dimensional Printers Result in Improved Trainee Anatomic Understanding. Urology. 2015;85:1257-1261. Komai Y, Sugimoto M, Gotohda N, et al. Patient-specific 3-dimensional Printed Kidney Designed for "4D" Surgical Navigation: A Novel Aid to Facilitate Minimally Invasive Offclamp Partial Nephrectomy in Complex Tumor Cases. Urology. 2016;91:226-233. Maddox MM, Feibus A, Liu J, Wang J, Thomas R, Silberstein JL. 3D-printed soft-tissue physical models of renal malignancies for individualized surgical simulation: a feasibility study. Journal of robotic surgery. 2017.
Figure 1. Axial CT image (A-C), 3D reconstruction (D-F, dV-Trainer simulation (G-I), and 3D printed physical models (J-L) for Model 1, 2, and 3.
M
AN US
CR IP T
ACCEPTED MANUSCRIPT
AC
CE
PT
ED
Figure 2. Histogram representing distribution of total tumor localization score averaged for all 3 models between subjects in the dV-Trainer and No-dV-Trainer groups. A lower average score demonstrates greater precision.
ACCEPTED MANUSCRIPT
Table 1. Participant demographics. Participant demographics
Table 1.
Age
23.5 (1.9)
dvTrainer 50 23.8 (2.5)
P value 0.528
CR IP T
No. participants
No dVTrainer 50
MS Year
0.600
MS1 MS2 MS3 MS4 Future Specialty
24 (48) 19 (38) 5 (10) 2 (4)
20 (40) 21 (42) 4 (8) 5 (10)
26 (52) 1 (2) 23 (46)
29 (58) 2 (4) 19 (38)
0.604
AN US
Surgical Non-Surgical Undecided dv-Trainer Warm-Up
Targeting
929 (231)
Pick and Place
651 (220)
ED
M
3-Dimensional Aptitude Score (0 to 1; 1 is best) All data presented as means (st. dev) or # per group (%)
0.87 (0.18)
922 (222) 727 (265) 0.83 (0.20)
0.887 0.120 0.225
AC
CE
All Models
PT
Table 2. Model performance.
Distance Score Diameter (R.) Endophytic/Exophytic (E.) Nearness to Collecting System (N.) Anterior/Posterior (A.) Location Relative to Polar Lines (L.)
No dVTrainer
dVTrainer
P value
0.38 (0.35) 136 (90.7) 80 (53.3) 20 (13.3) 130 (86.7) 71 (47.3)
0.24 (0.29) 117 (78) 54 (36) 20 (13.3) 103 (68.7) 102 (68)
<0.001 0.004 0.004 1.000 <0.001 <0.001
Model 1
Distance Score Diameter (R.) Endophytic/Exophytic (E.) Nearness to Collecting System (N.) Anterior/Posterior (A.)
0.49 (0.32) 47 (94) 23 (46) 5 (10) 45 (90)
0.44 (0.26) 46 (92) 9 (18) 6 (12) 32 (64)
0.431 1.000 0.005 1.000 0.004
ACCEPTED MANUSCRIPT
Location Relative to Polar Lines (L.)
30 (60)
42 (84)
0.014
Model 2
Model 3
0.34 (0.39) 49 (98) 43 (86) 7 (14) 40 (80) 21 (42)
0.12 (0.28) 36 (72) 30 (60) 8 (16) 36 (72) 31 (62)
0.012 0.356 1.000 0.774 0.023 0.072
0.001 <0.001 0.006 1.000 0.483 0.045
M
AN US
Distance Score Diameter (R.) Endophytic/Exophytic (E.) Nearness to Collecting System (N.) Anterior/Posterior (A.) Location Relative to Polar Lines (L.) All data presented as means (st. dev) or # correct (%)
40 (80) 14 (28) 8 (16) 45 (90) 20 (40)
0.17 (0.23) 35 (70) 15 (30) 6 (12) 35 (70) 29 (58)
0.31 (0.31)
CR IP T
Distance Score Diameter (R.) Endophytic/Exophytic (E.) Nearness to Collecting System (N.) Anterior/Posterior (A.) Location Relative to Polar Lines (L.)
ED
Supplemental Figure 1. Study flow diagram. This process was repeated for each model (n=3).
PT
Supplemental Figure 2. 9 variants from which subjects were required to determine the correct tumor location for model #2. Supplemental Figure 3. Pre-study demographic and 3D Aptitude Questionnaire.
CE
Supplemental Figure 4. Post-model questionnaire (same for Models 1-3).
AC
Supplemental Video. Video of novel edge detection algorithm to generate renal models from CT imaging.
Patient-specific virtual simulation - a state of the art approach to teach renal tumor localization Arun Rai , Jason M. Scovell , Ang Xu , Adithya Balasubramanian , Ryan Siller , Taylor Kohn , Young Moon , Naveen Yadav , Richard E. Link PII: DOI: Reference:
S0090-4295(18)30540-5 10.1016/j.urology.2018.04.043 URL 21063
To appear in:
Urology
Received date: Revised date: Accepted date:
7 January 2018 19 April 2018 27 April 2018
Please cite this article as: Arun Rai , Jason M. Scovell , Ang Xu , Adithya Balasubramanian , Ryan Siller , Taylor Kohn , Young Moon , Naveen Yadav , Richard E. Link , Patient-specific virtual simulation - a state of the art approach to teach renal tumor localization, Urology (2018), doi: 10.1016/j.urology.2018.04.043
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
Patient-specific virtual simulation - a state of the art approach to teach renal tumor localization Arun Rai1* and Jason M. Scovell1,2*, Ang Xu1, Adithya Balasubramanian1, Ryan Siller1, Taylor Kohn1, Young Moon1, Naveen Yadav1, Richard E. Link1,3 1
Scott Department of Urology, 2Translational Biology and Molecular Medicine and 3Dan
USA. *AR and JMS contributed equally to this work
AN US
The authors declare no conflict of interest Abstract word count: 249 Manuscript word count: 2929 Corresponding author: Richard E. Link
CR IP T
L. Duncan Comprehensive Cancer Center, Baylor College of Medicine, Houston, TX,
PT
ED
M
Richard E. Link, MD, PhD Carlton-Smith Endowed Chair in Urologic Education Associate Professor of Urology Scott Department of Urology Baylor College of Medicine 7200 Cambridge, MC BCM380, A10.107 Houston, TX 77030, U.S.A. Tel: (713) 798-7670 Fax: (713) 798-5553 Email: [email protected]
Key Words: education, algorithm, 3d printing, robotic surgery, partial nephrectomy
CE
Running title: Patient-specific virtual simulation - a state of the art approach to teach renal tumor localization
AC
Source of Funding: Research reported in this publication was supported by the National Institute of General Medical Sciences of the National Institutes of Health under Award Number T32GM088129 as well as the loan of the dV-Trainer simulator platform from Mimics Technologies, Inc. (Seattle, WA). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health or Mimics Technologies.
ACCEPTED MANUSCRIPT
ABSTRACT OBJECTIVES: To test a novel visuospatial testing platform improve trainee ability to convert two-dimensional (2D) to three-dimensional (3D) space.
CR IP T
METHDOS: Medical students were recruited from Baylor College of Medicine and McGovern Medical School (Houston, TX). We 3D reconstructed three partial
nephrectomy cases using a novel, rapid and highly accurate edge-detection algorithm. Patient-specific reconstructions were imported into the dV-Trainer (Mimics
AN US
Technologies, Seattle, WA) and used to generate custom 3D printed physical models. Tumor location was altered digitally to generate nine physical models for each case, one with the correct tumor location and eight with sham locations. Subjects were randomized 1:1 into the dV-Trainer (intervention) and No-dV-Trainer (control) groups.
M
Each subject completed the following steps: (1) visualization of computed-tomographic
ED
images, (2) visualization of the reconstructed kidney and tumor in the dV-Trainer (intervention group only), and (3) selection of the correct tumor location on the 3D
PT
printed models (primary outcome). Normalized distances from the correct tumor location
CE
were quantified and compared between groups. RESULTS: 100 subjects were randomized for this study. dV-Trainer use significantly
AC
improved subjects ability to localize tumor position (Tumor Localization Score: 0.24 vs. 0.38, p<0.001). However, subjects in the No-dV-Trainer group more accurately assigned R.E.N.A.L. CONCLUSION: Even brief exposure to interactive patient-specific renal tumor models improves a novice’s ability to localize tumor location. Virtual reality simulation prior to
ACCEPTED MANUSCRIPT
surgery could benefit trainees learning to localize renal masses for minimally invasive partial nephrectomy.
CR IP T
INTRODUCTION
One of the major technical hurdles a novice faces early in surgical training in developing the ability to train their visuospatial ability. Visuospatial ability refers to the capacity for mentally visualize and/or manipulate objects (4). Many surgical diagnoses are made by
AN US
standard two dimensional (2D) imaging such as Computed Tomography (CT) and
intervention requires the surgeon to translate these findings into a three dimensional (3D) mental image. This challenge is further amplified during minimally-invasive surgery
M
such as laparoscopic or robotic surgery where anatomic evaluation is indirect (via camera) and there is a lack or complete absence of direct tactile feedback. Not
ED
surprisingly, a recent meta-analysis concluded that performance on visuospatial aptitude, and in particular dynamic three dimensional testing, correlates with
PT
laparoscopic skills.1
CE
To address the importance of visuospatial ability, we have developed a novel training platform that utilizes a dynamic 3D virtual simulation environment (dV-Trainer) as a
AC
training bridge between 2D and 3D space. This platform evaluates visuospatial recognition using 3D printed models developed from standard 2D imaging. We have chosen robotic-assisted laparoscopic partial nephrectomy (RALPN) as a surgical test case for our novel training platform to assess visuospatial understanding. RALPN is a challenging procedure to learn and visuospatial understanding of correct
ACCEPTED MANUSCRIPT
tumor position, orientation and relationship to vital renal structures is essential for successful tumor excision and renal repair.5, 6 Safe and successful RALPN requires a subset of specialized skills that can be challenging to learn: sophisticated tumor visualization within a compact three dimensional (3D) space, proper tissue handling
CR IP T
despite lack of haptic feedback, and execution of tumor resection under ischemiadependent time constraints.5
We hypothesized that manipulating a patient-specific tumor/kidney model within a 3D
AN US
virtual simulator (dV-Trainer) would improve a novice’s ability to localize tumor location on a physical model (visuospatial ability) as compared to interpreting standard 2 dimensional planar imaging alone.
M
SUBJECTS AND METHODS
ED
Subjects
After obtaining Institutional Review Board (IRB) approval, medical students (years 1 to
PT
4) were recruited as subjects from Baylor College of Medicine and University of Texas McGovern Medical School in Houston, TX during 2016. Model kidneys used for surgical
CE
simulation were generated from cross-sectional computed-tomographic (CT) imaging of three patients who had previously undergone RALPN with surgeon R.E.L. at our
AC
institution. These three models were selected based on their diverse tumor locations and differences in Nephrometry score.10 Image Reconstruction and Virtual Model Generation
ACCEPTED MANUSCRIPT
The three patient-specific 3D models were reconstructed based on the available preoperative CT imaging (Figure 1), and were converted into both a virtual reality model within the dV-Trainer environment (intervention) and a physical 3D printed plaster model (primary outcome). The 3D reconstruction was generated through a novel unpublished
CR IP T
―edge-detect‖ MATLAB algorithm (MathWorks; Natick, MA) that utilizes image
processing to detect the edges of the kidney. This is made possible by the sudden
gradation of contrast enhancement on nephrogenic phase of the CT between renal parenchyma and surrounding fat which is then capable of generating a series of ―edges‖
AN US
through each axial CT section. This algorithm allows for rapid reconstruction of kidney and tumor in less than 60 seconds (Supplemental Video 1). Upon generating a series of edges for the entirety of the kidney, edges are retraced with CV curves in a plugin for a
M
3D modeling program. Each image is a cross sectional curve in the program and are then logged into an accurate 3D model of the kidney. Tumors were identified and
ED
marked in a similar fashion. The resulting in silico 3D model was saved in .stl file format and exported for final processing, refinement and importation into the dv-Trainer system
PT
(Mimic Technologies Inc., Seattle WA). This process is the first, to our knowledge, to be
CE
used to efficiently generate high-fidelity 3D reconstructions in an automated fashion. Upon generation of the 3D virtual reality model by Mimic Technologies (Seattle, WA),
AC
proofs underwent final approval for face validity with the authors prior to study initiation. Orientation markers were included in the dV-Trainer operating field for cranial, caudal, anterior and posterior. Virtual reality models could be rotated about the hilum and manipulated in an anterior/posterior fashion. Transparency of the renal tissue could be toggled on the desktop unit adjacent to the dV-Trainer console. The transparency toggle
ACCEPTED MANUSCRIPT
was used to help identify the location of the renal tumor within the substance of the kidney parenchyma. 3D ColorJet Printed Model Generation
CR IP T
Similar to the virtual reality models, the 3D printed models were generated based on the same 3D reconstructions generated via the ―edge-detect‖ algorithm. These models were 3D printed using ColorJet technology (ProJet 260C, 3D Systems, Rock Hill, SC) with the parenchyma as red and the tumor as green to help subjects delineate the mass
AN US
from the kidney. All plaster models were digitally bivalved in the sagittal plane prior to printing in order to help delineate anterior/posterior as well as endophytic/exophytic orientation. For each of the three renal models, 1 ―true‖ patient-specific model was generated in which the renal mass was located in the orthotopic position as delineated
M
from the CT imaging. For each patient-specific model, 8 additional variants were
ED
generated wherein the renal mass was localized to other randomized locations in the parenchyma. False were placed throughout the kidney – some tumors were placed very
PT
close (<1 cm away) and others were placed at distant locations (opposite pole). Upon the generation of these models, all tumor locations were assigned a tumor localization
CE
score (TLS) on a scale from 0 (center of the correct renal mass, orthotopic position) to 1
AC
(center of the furthest renal mass as measured as a straight line (i.e. if the furthest mass was noted to be 5 cm, a model with a tumor located 2.5 cm away was assigned a score of 0.5). Normalization was performed to allow quantification and enable comparisons across models. Study Design
ACCEPTED MANUSCRIPT
Prior to study initiation, all participants were asked to complete a questionnaire which included questions about demographics, post-graduate plans (surgical, non-surgical, or undecided), as well as three visuospatial aptitude questions in order to assess baseline 3D spatial reasoning (Supplemental Figure 3). The study was carried out in multiple
CR IP T
small groups in the Baylor College of Medicine Simulation Lab in order to minimize wait time and ensure study fluidity given the large number of participants. Informed consent was obtained from each participant in accordance with our IRB protocol. A short
orientation presentation was provided to participants detailing the components of the
AN US
study, how to interpret CT imaging of the kidney and a review of how to calculate the R.E.N.A.L. Nephrometry score.10 After this point, all participants were taken to the dVTrainer and completed two orientation exercises (Camera Targeting and Pick and
M
Place) in order to familiarize participants with the dV-Trainer as well as to collect user metrics. At this time, participants were randomized 1:1 into two groups – ―No dV-
ED
Trainer‖ and ―dV-Trainer‖ based on an A:B pattern. Both groups began first with Model #1 and were allowed 5 minutes to review the CT imaging (Supplemental Figure 1).
PT
Upon review of the CT imaging, the ―dV-Trainer‖ group was then allowed to review and
CE
manipulate the renal model in the dV-Trainer virtual reality trainer for 5 minutes. In comparison, the ―no dV-Trainer‖ group was taken immediately after randomization to a
AC
workstation wherein the 9 plaster models were laid out (Supplemental Figure 2). Five minutes were allotted to the participants to select the model they felt best represented the renal tumor visualized on CT imaging. After completion of 5 minutes on the dVTrainer, the ―dV-Trainer‖ group was taken to the same workstation to select the most representative physical model. After each model, all participants were then asked to fill
ACCEPTED MANUSCRIPT
out a questionnaire regarding the tumor’s Nephrometry score as well as which physical model best represented the mass they had viewed on CT imaging (primary outcome) (Supplemental Figure 4). This process was repeated for models 2 and 3.
CR IP T
Statistical Analysis A power analysis was performed a priori to detect a medium effect size (Cohen d = 0.6) using a Student’s t-test (β = 0.20, α = 0.05) and determined that a total of 45 subjects were required per group for a total of 90 subjects. The primary outcome for this study
AN US
was the tumor localization score between subjects in the ―dV-Trainer‖ and ―No dV-
Trainer‖ groups. Comparisons of proportions were performed using the Chi-Square test and Fisher test where appropriate. We compared continuous variables using the Student’s t-test. Univariate analysis was performed using the Generalized Linear Model
M
with tumor localization score as the dependent variable. Statistical analysis was
ED
performed using R v.3.3 on RStudio v.1.0 and SPSS v22 (IBM, NY).
PT
RESULTS
Three models were generated for the study and graded by Nephrometry score by R.E.L.
CE
(Model 1: 1+2+3+a+1 = 7a, left sided; Model 2: 1+2+3+p+3 = 9p, left sided; Model 3: 1+1+3+p+2 = 7p, right sided). 100 medical students were recruited and participated in
AC
the study. Subjects were randomized to the intervention group (dV-Trainer; n = 50), or the control group (No dV-Trainer; n = 50). Between groups, there was no difference in subject age (dV: 23.5 vs. No-dV: 23.8, p = 0.53), MS year (p = 0.60), and desired future specialty (p = 0.60) (Table 1). We assessed spatial reasoning using a 3-question spatial aptitude test and two warm-up exercises on the dV-Trainer (targeting, pick and place).
ACCEPTED MANUSCRIPT
There was no baseline difference in 3D aptitude between groups (p = 0.23) or ability with the dV-Trainer (p = 0.89, p = 0.12). Subjects in the dV-Trainer group demonstrated improved tumor localization compared
CR IP T
to subjects that did not use the dV-Trainer (Tumor Localization Score: 0.24 vs. 0.38, p<0.001) (Figure 2, Table 2). There was no difference in tumor localization for Model 1 (0.44 vs 0.49, p=0.43), but significant differences were observed for Model 2 (0.17 vs. 0.31, p=0.01) and Model 3 (0.12 vs. 0.34, p=0.001).
AN US
Contrary to our observation that dV-Trainer use improved subject ability to localize tumor location, subjects that did not use the dV-Trainer more accurately assigned R.E.N.A.L. Nephrometry scores (56% vs. 63%, p = 0.013). Subjects in the No-dV-
Trainer group more accurately assigned the correct diameter (78% vs. 91%, p = 0.004),
M
if the mass was endophytic/exophytic (36% vs. 53%, p = 0.004), and if the mass was
ED
anterior or posterior (87% vs. 69%, p < 0.001). Subjects in the dV-Trainer group were better able to assign location of the mass relative to the polar lines (68% vs. 47%, p <
PT
0.001). There was no difference in capacity to determine nearness to the collecting
CE
system with or without dV-Trainer use (31% vs. 35%, p = 0.46). We performed a univariate analysis to determine if any additional factors may associate
AC
with ability to identify the correct physical tumor location (Supplemental Table 1). We found that only dV-Trainer use was associated with improved tumor localization (β = 0.137, 95% CI: 0.070 – 0.203, p < 0.001). Age, 3D aptitude score, dV-Trainer warm-up scores, ability to assign the correct R.E.N.A.L. Nephrometry score, MS year, and desired future specialty did not associate with tumor localizing ability.
ACCEPTED MANUSCRIPT
DISCUSSION Surgical simulation can be divided into two broad categories – mechanical and virtualreality based simulation. Mechanical simulators include simple box trainers wherein
CR IP T
tasks are completed via videoscopic guidance. In virtual reality simulators, tasks are completed in a computer-generated environment often with 3D visualization.6 These virtual reality simulators have been developed for both robotic and traditional
laparoscopic kidney surgery. However, the validity of these simulators for improving
AN US
surgical performance remains controversial.
We hypothesized that even a short exposure to an interactive 3D virtual reality simulator (dV-Trainer) containing a patient-specific model could enhance a novice’s ability to create an accurate mental picture of tumor location. We tested this hypothesis by
M
asking the subject to identify the correct tumor location from a panel of different physical
ED
models containing variations in tumor location. We chose to recruit medical students to test this hypothesis due to their relatively uniform novice status with respect to
PT
visualizing renal masses from 2D planar images. The pool of medical students also provided the large number necessary to adequately power this study. Here we
CE
demonstrate that novice trainees who utilize the dV-Trainer virtual environment in
AC
addition to standard 2D planar imaging more accurately localize tumors in 3D physical models as compared to trainees that have access to 2D planar images alone. The R.E.N.A.L. Nephrometry score was developed to stratify the complexity of renal tumors based on several measurements of tumor anatomy (R—radius, E—exophytic vs endophytic, N—nearness of tumor to collecting system, A—anterior vs posterior, L—
ACCEPTED MANUSCRIPT
location relative to polar lines).10 Several studies have validated the utility nephrometry scoring in predicting perioperative and renal functional outcomes.13 More recent studies have demonstrated that radius, endophytic location and nearness to the collecting system are specifically associated with significant changes in warm ischemia time and
CR IP T
overall operative time.
We anticipated that access to the dV-Trainer would also enhance a novice trainee’s ability to accurately assign R.E.N.A.L. nephrometry scores. However, this was not
AN US
uniformly true. Despite improving the novice’s ability to localize a tumor on a 3D
physical model, use of the dV-Trainer decreased the accuracy with which subjects could attribute tumor diameter, how endophytic/exophytic, and how anterior or posterior the mass was for nephrometry scoring. Use of the dV-Trainer only increased subject ability
M
to properly localize where the tumor fell within polar lines. Perhaps, the apparent discrepancy between ability to localize a tumor on a physical model and assign the
ED
correct proper nephrometry score underlies the inherent complexity of converting the
PT
objective description obtained from 2D planar imaging to the real anatomical orientation and position of the tumor and mass. Certain aspects of the nephrometry score,
CE
specifically the nearness to the collecting system, require careful review and measurement on planar images that may require more than 5 minutes of study. This is
AC
one potential explaination for some of the reported discrepancies between nephrometry scores from urologists and radiologists for the same tumor case.14 We suspect that nephrometry scoring is a more challenging and time consuming exercise for novices than simply developing an accurate mental image of tumor location sufficient for surgical performance.
ACCEPTED MANUSCRIPT
3D printing has become more accessible, which is driving development of patientspecific physical models as surgical training tools. Knoedler et al. demonstrated that 3D printed physical models can enhance medical students’ ability to correctly assign the R, N, and L components of the Nephrometry score compared to standard 2D planar
CR IP T
imaging alone.15, 16 This highlights the potential educational differences between
physical and digital models, at least for novice trainees. Several groups, including our own, have also reported development of patient-specific soft-tissue models that can be utilized for pre-surgical rehearsal, or as part of a patient-specific ex vivo training
AN US
program.8, 17, 18 Clearly, these exciting new technologies have great potential to enhance surgical education and further randomized trials are needed to define the value of each type of model.
M
This study has several important strengths and weaknesses that warrant discussion. Our conclusions are strengthened by the prospective double-bind randomized study
ED
design and by sufficient accrual to satisfy our a priori power analysis. Our study also
PT
employed three relatively novel technologies: renal and tumor modeling using an automated point-and-click algorithm, importation of patient-specific models into a 3D
CE
dV-Trainer simulator, and 3D printing of digitally altered model variations. Our results support the hypothesis that even brief use of a 3D simulator can enhance a novice’s
AC
ability to identify renal tumor location in the physical world. Developing new technologies and simulators can be a costly and time-consuming enterprise that requires solid proof of concept to justify investment. By performing this study in a relatively homogeneous population of subjects, using a randomized approach, we intended to provide the best possible pilot data to assess the potential
ACCEPTED MANUSCRIPT
benefit of simulation in teaching tumor localization. Our intent was to determine whether or or not VR simulation can improve a novice’s visuospatial understanding of patient anaotmy. The physics and engineering for patient specific VR surgery does not exist yet and hence this study was designed and completed to help demonstrate the utility of VR
CR IP T
in its current state.
Academic urologists who teach trainees and all practicing urologists have a vested interest in how future trainees are taught to perform robotic surgery. A very large
AN US
proportion of urologic surgeons participate in teaching urology trainees. Many senior urologic surgeons who have only recently adopted robotics for partial nephrectomy likely struggle with the same visualization challenges faced by trainees when they switch from an open to a robotic environment for this procedure. Minimally invasive
M
surgery provides new opportunities for innovation in designing training regimens. Many of these innovative new approaches, whether they involve surgical simulation, the use
ED
of video games to teach surgery, 3D printed anatomic models, laparoscopic trainer
PT
boxes, and video based surgical evaluation are still developing. It is important for the major stakeholders to continue to explore, debate, and refine these methods as these
CE
techniques will drive how future urologists are trained.
AC
There are several important limitations to this study, including the novice study population (medical students) and relatively low-complexity nature of our models. While medical students are an excellent choice for a population of RALPN novices, these conclusions may not be translatable to a more experienced trainee cohort. Although we provided a standardized tutorial on how to interpret renal CT images prior to this study, it is difficult to determine how effective this tutorial was in establishing an imaging
ACCEPTED MANUSCRIPT
interpretation baseline for medical students. We suspect that more advanced trainees (i.e. surgical interns) would still benefit from using the dV-Trainer to localize renal masses; however, this effect may not be as substantial as we observed in the medical student population. Our models were generated using a novel point-and-click algorithm
CR IP T
which enabled rapid generation of digital models. This approach reduces a significant time-dependent barrier required for patient-specific surgical simulation. Although these models are visually similar to models generated by the gold-standard manual
segmentation technique, we did not rigorously quantify their similarity here. This is an
AN US
intended area of future research. These models represented only the patient’s renal parenchyma and tumor. In the dV-Trainer virtual reality atmosphere we added standard (non-patient specific) vasculature and ureter. Although difficult to determine why a
M
difference was noted in models 2 and 3 but not in model 1, we speculate that this was due to the random placement of a ―false‖ tumor very close to the ―true‖ location. As a
ED
result, participants in both groups struggled to differentiate between these two locations. Generating higher fidelity models that include perinephric fat, the renal collecting
PT
system, and adjacent abdominal organs may be beneficial when translating this
CE
application to patient-specific pre-surgical training.
AC
CONCLUSION
Visualizing patient-specific kidneys and tumors within the dV-Trainer 3D surgical simulation environment, even for only 5 minutes, improves a novice’s ability to localize correct tumor location in the physical world. Visual processing of 2D imaging into 3D anatomical structures is a challenging step for many surgeons. Use of the dV-Trainer as
ACCEPTED MANUSCRIPT
a mechanism to test and improve this ability may provide a powerful new approach for novices to adapt this skill towards a broad array of surgical procedures. REFERENCES
8. 9. 10. 11. 12.
AC
13.
AN US
7.
M
6.
ED
5.
PT
2. 3. 4.
Kramp KH, van Det MJ, Hoff C, Veeger NJ, ten Cate Hoedemaker HO, Pierie JP. The predictive value of aptitude assessment in laparoscopic surgery: a meta-analysis. Med Educ. 2016;50:409-427. SEER. Cancer Stat Facts: Kidney and Renal Pelvis Cancer2016. ACS. Cancer Facts & Figures 20172017. Luciani LG, Cestari R, Tallarigo C. Incidental renal cell carcinoma-age and stage characterization and clinical implications: study of 1092 patients (1982-1997). Urology. 2000;56:58-62. Ahmadi H, Liu JJ. 3-D Imaging and Simulation for Nephron Sparing Surgical Training. Current urology reports. 2016;17:58. Albani JM, Lee DI. Virtual reality-assisted robotic surgery simulation. Journal of endourology. 2007;21:285-287. Patel HD, Mullins JK, Pierorazio PM, et al. Trends in renal surgery: robotic technology is associated with increased use of partial nephrectomy. The Journal of urology. 2013;189:1229-1235. von Rundstedt FC, Scovell JM, Agrawal S, Zaneveld J, Link RE. Utility of patient-specific silicone renal models for planning and rehearsal of complex tumour resections prior to robot-assisted laparoscopic partial nephrectomy. BJU international. 2016. von Rundstedt FC, Scovell JM, Agrawal S, Zaneveld J, Link RE. Utility of patient-specific silicone renal models for planning and rehearsal of complex tumour resections prior to robot-assisted laparoscopic partial nephrectomy. BJU international. 2017;119:598-604. Kutikov A, Uzzo RG. The R.E.N.A.L. nephrometry score: a comprehensive standardized system for quantitating renal tumor size, location and depth. The Journal of urology. 2009;182:844-853. Leow JJ, Heah NH, Chang SL, Chong YL, Png KS. Outcomes of Robotic versus Laparoscopic Partial Nephrectomy: an Updated Meta-Analysis of 4,919 Patients. The Journal of urology. 2016;196:1371-1377. Choi JE, You JH, Kim DK, Rha KH, Lee SH. Comparison of perioperative outcomes between robotic and laparoscopic partial nephrectomy: a systematic review and meta-analysis. European urology. 2015;67:891-901. Watts KL, Ghosh P, Stein S, Ghavamian R. Value of Nephrometry Score Constituents on Perioperative Outcomes and Split Renal Function in Patients Undergoing Minimally Invasive Partial Nephrectomy. Urology. 2017;99:112-117. Benadiba S, Verin AL, Pignot G, et al. Are urologists and radiologists equally effective in determining the RENAL Nephrometry score? Annals of surgical oncology. 2015;22:16181624. Silberstein JL, Maddox MM, Dorsey P, Feibus A, Thomas R, Lee BR. Physical models of renal malignancies using standard cross-sectional imaging and 3-dimensional printers: a pilot study. Urology. 2014;84:268-272.
CE
1.
CR IP T
REFERENCES
14. 15.
ACCEPTED MANUSCRIPT
16. 17.
AC
CE
PT
ED
M
AN US
CR IP T
18.
Knoedler M, Feibus AH, Lange A, et al. Individualized Physical 3-dimensional Kidney Tumor Models Constructed From 3-dimensional Printers Result in Improved Trainee Anatomic Understanding. Urology. 2015;85:1257-1261. Komai Y, Sugimoto M, Gotohda N, et al. Patient-specific 3-dimensional Printed Kidney Designed for "4D" Surgical Navigation: A Novel Aid to Facilitate Minimally Invasive Offclamp Partial Nephrectomy in Complex Tumor Cases. Urology. 2016;91:226-233. Maddox MM, Feibus A, Liu J, Wang J, Thomas R, Silberstein JL. 3D-printed soft-tissue physical models of renal malignancies for individualized surgical simulation: a feasibility study. Journal of robotic surgery. 2017.
Figure 1. Axial CT image (A-C), 3D reconstruction (D-F, dV-Trainer simulation (G-I), and 3D printed physical models (J-L) for Model 1, 2, and 3.
M
AN US
CR IP T
ACCEPTED MANUSCRIPT
AC
CE
PT
ED
Figure 2. Histogram representing distribution of total tumor localization score averaged for all 3 models between subjects in the dV-Trainer and No-dV-Trainer groups. A lower average score demonstrates greater precision.
ACCEPTED MANUSCRIPT
Table 1. Participant demographics. Participant demographics
Table 1.
Age
23.5 (1.9)
dvTrainer 50 23.8 (2.5)
P value 0.528
CR IP T
No. participants
No dVTrainer 50
MS Year
0.600
MS1 MS2 MS3 MS4 Future Specialty
24 (48) 19 (38) 5 (10) 2 (4)
20 (40) 21 (42) 4 (8) 5 (10)
26 (52) 1 (2) 23 (46)
29 (58) 2 (4) 19 (38)
0.604
AN US
Surgical Non-Surgical Undecided dv-Trainer Warm-Up
Targeting
929 (231)
Pick and Place
651 (220)
ED
M
3-Dimensional Aptitude Score (0 to 1; 1 is best) All data presented as means (st. dev) or # per group (%)
0.87 (0.18)
922 (222) 727 (265) 0.83 (0.20)
0.887 0.120 0.225
AC
CE
All Models
PT
Table 2. Model performance.
Distance Score Diameter (R.) Endophytic/Exophytic (E.) Nearness to Collecting System (N.) Anterior/Posterior (A.) Location Relative to Polar Lines (L.)
No dVTrainer
dVTrainer
P value
0.38 (0.35) 136 (90.7) 80 (53.3) 20 (13.3) 130 (86.7) 71 (47.3)
0.24 (0.29) 117 (78) 54 (36) 20 (13.3) 103 (68.7) 102 (68)
<0.001 0.004 0.004 1.000 <0.001 <0.001
Model 1
Distance Score Diameter (R.) Endophytic/Exophytic (E.) Nearness to Collecting System (N.) Anterior/Posterior (A.)
0.49 (0.32) 47 (94) 23 (46) 5 (10) 45 (90)
0.44 (0.26) 46 (92) 9 (18) 6 (12) 32 (64)
0.431 1.000 0.005 1.000 0.004
ACCEPTED MANUSCRIPT
Location Relative to Polar Lines (L.)
30 (60)
42 (84)
0.014
Model 2
Model 3
0.34 (0.39) 49 (98) 43 (86) 7 (14) 40 (80) 21 (42)
0.12 (0.28) 36 (72) 30 (60) 8 (16) 36 (72) 31 (62)
0.012 0.356 1.000 0.774 0.023 0.072
0.001 <0.001 0.006 1.000 0.483 0.045
M
AN US
Distance Score Diameter (R.) Endophytic/Exophytic (E.) Nearness to Collecting System (N.) Anterior/Posterior (A.) Location Relative to Polar Lines (L.) All data presented as means (st. dev) or # correct (%)
40 (80) 14 (28) 8 (16) 45 (90) 20 (40)
0.17 (0.23) 35 (70) 15 (30) 6 (12) 35 (70) 29 (58)
0.31 (0.31)
CR IP T
Distance Score Diameter (R.) Endophytic/Exophytic (E.) Nearness to Collecting System (N.) Anterior/Posterior (A.) Location Relative to Polar Lines (L.)
ED
Supplemental Figure 1. Study flow diagram. This process was repeated for each model (n=3).
PT
Supplemental Figure 2. 9 variants from which subjects were required to determine the correct tumor location for model #2. Supplemental Figure 3. Pre-study demographic and 3D Aptitude Questionnaire.
CE
Supplemental Figure 4. Post-model questionnaire (same for Models 1-3).
AC
Supplemental Video. Video of novel edge detection algorithm to generate renal models from CT imaging.