A Robotic System for Implant Modification in Single-stage Cranioplasty
AA Robotic System for Implant Modification in Single-stage Cranioplasty
Shuya Liu ∗ , Wei-Lun Huang (cid:63) , Chad Gordon † and Mehran Armand ∗ , † Abstract — Craniomaxillofacial reconstruction with patient-specific customized craniofacial implants (CCIs) is most com-monly performed for large-sized skeletal defects. Because theexact size of skull resection may not be known prior to thesurgery, in the single-stage cranioplasty, a large CCI is pre-fabricated and resized intraoperatively with a manual-cuttingprocess provided by a surgeon. The manual resizing, however,may be inaccurate and significantly add to the operating time.This paper introduces a fast and non-contact approach forintraoperatively determining the exact contour of the skullresection and automatically resizing the implant to fit theresection area. Our approach includes four steps: First, apatient’s defect information is acquired by a 3D scanner.Second, the scanned defect is aligned to the CCI by registeringthe scanned defect to the reconstructed CT model. Third, acutting toolpath is generated from the contour of the scanneddefect. Lastly, the large CCI is resized by a cutting robot to fitthe resection area according to the given toolpath. To evaluatethe resizing performance of our method, six different resectionshapes were used in the cutting experiments. We compared theperformance of our method to the performances of surgeon’smanual resizing and an existing technique which collects thedefect contour with an optical tracking system and projectsthe contour on the CCI to guide the manual modification.The results show that our proposed method improves theresizing accuracy by 56 % compared to the surgeon’s manualmodification and 42 % compared to the projection method. I. I
NTRODUCTION
Cranioplasty is a procedure to treat cranial defects dueto trauma, injury, or neurosurgical procedures for braintumors, aneurysms or epilepsy [1]. Conventional cranioplastyis a two-stage process that repairs skull deformities in adelayed operation [2]. Such process requires the skull tobe partially removed from the patient who then has to waitfor the design and fabrication of the replacing implant forthree to four weeks. In contrast, single-stage cranioplastyaims to restore aesthetic appearance immediately followingcraniectomy within one single operation, therefore, decreas-ing operative times and speeding up the patient’s recovery[3]. In practice, cranioplasty replaces the skull defect withan alloplastic implant instead of using a patient’s autologousskull-bone [4]. ∗ Shuya Liu is with the Department of Mechanical Engineering, JohnsHopkins University, Baltimore, MD. [email protected] (cid:63)
Weilun Huang is with the Department of Computer Science, JohnsHopkins University, Baltimore, MD. [email protected] † Chad Gordon is with the Neuroplastic and ReconstructiveSurgery, Johns Hopkins School of Medicine, Baltimore, MD. [email protected] ∗ † Mehran Armand is with the Department of Mechanical Engineer-ing, Johns Hopkins University, Baltimore, MD; the Department of Or-thopaedic surgery, the Johns Hopkins School of Medicine, Baltimore, MD. [email protected]
Fig. 1. Top: A clinical example of the single-stage cranioplasty with aprefabricated large CCI. i) the surgeon marks the defect contour on the CCI.ii) the implant is manually modified by a surgical cutter. iii) the resized CCIis placed to the skull defect. Middle: The workflow of robotic single-stagecranioplasty. i) 3D scanning generates a scanned model of the defect skull.ii) the scanned model is registered to its CT model. iii) a cutting toolpath isgenerated. iv) The implant is resized by the robot. Bottom: Left: The plasticdefect skull is scanned by a handheld 3D scanner. Right: robot resizes theoversized implant.
Several approaches are utilized to generate CCIs. Moldingtechnique has been applied to form CCIs in the operat-ing room [5]. This method requires injecting liquid bio-compatible materials such as poly-methyl-methacrylate di-rectly into the defect of the skull or a molding templategenerated using the autologous bones. However, moldinga CCI directly on the defect may release an exothermicreaction damaging nearby tissue [6]. Moreover, autologousbones cannot always be used to create a negative imprintand are limited in their ability to eliminate the discontinuitiesbetween the boundaries. Another commonly used approachis a cutting guide, in which a customized implant anda cutting guide are prefabricated and the surgeon resectsthe patient’s skull along the cutting guide so that skulldefect can be immediately closed with the prefabricated CCI[7], [8]. In addition, some other groups considered usingoptical navigation systems to achieve planned resections.[9], [10]. Although these methods are capable of repairingskull deformities within one operation, they do not consider a r X i v : . [ c s . R O ] J a n he possibilities for intraoperative plan changes, limitingneurosurgeons’ flexibility in reaching specific regions of thebrain. In [3], [11], a clinical approach using prefabricatedlarge CCIs in single-stage cranioplasty is presented. Thisapproach requires a surgeon to intraoperatively modify anoversized CCI by manually resizing it, which is often poorin accuracy and time-consuming.Computer-assisted single-stage cranioplasty provides amethod to help surgeons better visualize the defect contourby directly projecting the defect contour on an oversizedCCI. This method utilizes an optical tracking system tocollect data points of the defect contour [12]. However,this system is difficult to set up and requires line-of-sight.Moreover, the planar projection from a fixed configurationmay not be suitable for implants with complex structures.To address the above-mentioned problems, we previouslydeveloped a portable projection mapping device that trackssurgical instruments and projects a 3D defect contour ontothe implant in real-time from any angle without informationloss [13]. Although this approach improves the accuracy ofprojection mapping for medical augmented reality, it canonly collect one data point per frame with a digitizinginstrument. Therefore, this approach takes longer to collectsufficient data points for registration.Recent advances in 3D scanning technologies provide newvenues for extending the application of medical robots in theoperating room [14]. 3D scanning generates high-precision3D models of real-world objects that can be recognizedwithin a robot’s workspace to achieve specific autonomoustasks. Different from the optical tracking system, a 3Dscanner can collect thousands of data points per framewithout contacting the object. The use of a 3D scanner forskull defect reconstruction can simplify and expedite theidentification of the defect’s contour.In this paper, we present a novel system for generatingprecise CCIs for patients in single-stage cranioplasty. Thesystem consists of a 3D scanner and a cutting robotic arm.The 3D scanning technique enables fast registration andgeneration of cutting toolpaths. The cutting robotic armprovides stable and accurate performance compared to themanual cutting approach by surgeons. The contributions ofthis work include: 1) We proposed a fast and non-contactapproach for acquiring defect contour information using ahandheld 3D scanner. 2) We developed an algorithm for gen-erating cutting toolpaths from the extracted defect contours.3) We integrated a robotic system for automatic implantmodification. 4) Our approach improved the accuracy of theimplant modification compared to the conventional manualapproach and an existing method using an optical trackingsystem. II. M ETHOD
We developed a robotic system for resizing oversized CCIsduring intraoperative operation. The system consists of ahandheld Artec Space Spider 3D scanner (up to 0.1 mmresolution) and a KUKA LBR iiwa 7 R800 robotic arm. The3D scanner was utilized to acquire 3D information of the
Fig. 2. Coordinate transformations between different models. Left: the 3dscanned defect mesh model in F scan is registered to the CT model in F CT .Right: the bottom (blue) is a prefabricated oversized CCI with a referenceframe F ref defined by three spherical markers. F Base is the robot’s baseframe. F ee is the robot’s end-effector frame. F TCP is the frame attached tothe calibrated TCP. The transformation between different coordinate framesare shown as T . skull defect and to export a refined mesh (Fig. 1, bottom,left). The KUKA robotic arm was modified into a cuttingworkstation by adding a spindle tool to the robot’s end-effector (Fig. 1, bottom right).Our CCI generation method includes four steps (Fig. 1,middle): The information of a defected skull is collectedusing a 3D scanner. The scanned data is registered tothe CT model space. A cutting toolpath is generated byextracting the defect contour. A cutting robot resizes theCCI according to the generated toolpath.
A. 3D reconstruction of a patient’s defect skull
A 3D scanning process was first implemented using ahandheld 3D scanner. During this process, the 3D scannerwas held by hand at an approximate half meter distance fromthe skull and moved slowly around it. This process could beterminated when there were sufficient 3D point cloud datashown in the visualization software. This process usuallytakes less than two minutes.
B. Patient-CT registration
The 3D-scanned data was then registered to the preopera-tive CT model of the patient’s skull. This process transformedthe scanned data to the CT model space. An iterativeclosest points (ICP) registration method was applied with aninitialization using anatomical points [15]. The anatomicalpoints were selected from the scanned model of the patient.During the registration process, the 3D-scanned data tendsto mistakenly overlap with the inner layer of the skull,because of the similar geometric feature between the innerand outer layer. To prevent this problem, we designed apreprocessing algorithm to remove the CT model’s innerlayer and to generate a polygon surface model.In this method, we defined the 3D position of each vertexin the CT skull mesh as q i ∈ R . Then the center of themesh o ∈ R can be approximated by: o = ∑ ni = q i n . Thenwe constructed vector v i , which points from the center o ig. 3. Toolpath generation. (i) The raw 3D-scanned defect is filtered bycurvature. The largest connected component is preserved. (ii) The remainingvertices are fitted to a plane, transformed to a local cylindrical coordinatesystem defined on the plane parameterized as ( θ , r , h ) , and then fitted to apolynomial curve. (iii) The fitted curve is turned into a spline interpolatedthrough control points. (iv) The control points are projected onto theimplant’s top surface. (v) A cutting toolpath is generated from the splinecurve. to each vertex q i . Since each vertex in the skull mesh is alsoassociated with a normal vector n i of its own. The vectors ofthe inner layer point to the hollowed space inside the skulltowards the center o , while the vectors of the outer layerpoint to the opposite directions. As a result, the sign of q i · n i determines whether this vertex is located in the inner layeror the outer layer. We removed all the vertices with non-positive inner products to ensure that only outer layer of theCT mesh was preserved. C. Toolpath Generation
To generate a toolpath for resizing process, the defectcontour was first extracted from the 3D-scanned mesh ofthe patient’s head. Then, the scanned defect was aligned tothe implant by registering the defect to the CT model. Atoolpath consisting cutting positions and vectors along theextracted contour was generated in the coordinate frame ofthe CT model (Fig. 3). We implemented the following stepsto generate 3D cutting toolpaths using Pyvista [16]:
1) Curvature Filter:
To extract the contour of the defect,a curvature filter was applied to the vertices of the scannedmesh and followed by manual adjustments. We utilized acurvature filter to determine the local mean curvature alongthe surface of the defect and extracted the high curvaturevalue above a designed threshold. The mean curvature H,was calculated as H = ( κ + κ ) , where κ and κ are themaximum and minimum values of the principal curvature onthe mesh [17]. This filter was able to identify crease changesin the curvature of the surface. After curvature filtering,only the largest component was kept (Fig. 3, i). Additionalmanual adjustments could further remove potential redundantvertices that were connected to the largest component.
2) Curve Fitting:
After removing the redundant vertices,a group of vertices around the defect contour were leftdenoted as M (Fig. 3, ii). We first defined a local cylindricalcoordinate system on a best-fit plane of the extracted contourparameterized as ( θ , r , h ) . The center of the cylindrical Fig. 4. Implant and skull defect generation. Top: the skull defect (a)and the implant (b) are generated by (1) a boolean operation between theskull and two customized contours (red contour: skull defect, blue contour:implant) and (2) attaching spherical markers to their surfaces. The skulldefect is cropped to a 3-D printable size. Bottom: the skull defects (top)and implants (bottom) for six different specimens are generated. coordinate system was obtained by the mean coordinates ofthe contour. Then a nonlinear least squares method was usedto fit a polynomial expression of the curve.
3) Spline Projection:
The fitted curve was then convertedinto a spline, which consists of control points along thefitted curve. Since the oversized implant and defect werealready aligned in the CT coordinate system after Patient-CT registration, the control points were projected onto thetop surface of the implant. Therefore, the shape of the splineremained consistent with the curvature of the implant (Fig.3, iv).
4) Toolpath Generation:
For each discretized point P i along the spline curve, a unit 3D vector V i was generatedto guide tool axis of the cutting path. At first, each V i wasinitialized as n o of the best-fit plane. By combining n o andeach vector t i from the center point O c to each discretizedpoint P i , the generated cutting vectors V i could be tilted by acut angle. The obtained vector V i were then assigned to eachdiscretized point P i to form the toolpath. To compensate thetool radius, the discretized points ∑ P i could be expandedwith an offset equal to the radius of the cutting tool (Fig. 3,v). III. E XPERIMENTAL S ETUP
To evaluate the implant-resizing accuracy of our integratedsystem, we compared our method with the surgeon’s manual-resizing method and an existing optical tracking method.We conducted six experiments with independently generatedskull defects with different sizes and shapes using booleanoperations. As shown in Fig. 4 (Top), we first subtractedthe mesh inside the red contour from a complete skull tocreate a defect on the skull. On the same complete skull, theimplant mesh inside the blue contour was extracted to createits corresponding oversized implant. The defected skull wasfurther cropped to a 3D printable size and was fabricatedusing a 3-D printer (Stratasys F370, ABS material). Finally,we attached three spherical markers on the top surfacef each cropped defect and its corresponding implant forPatient-CT registration and implant localization (Fig. 2, F re f ). A. Method compassion
We compared the implants generated by our method withmanual resizing method, as well as the optical trackingmethod used by [12]. For robotic cutting, the cut depth wasset to 3 mm, which is the same as thickness of the designedimplants. The cut angle was set to 20 degrees for all thegenerated cutting toolpaths.
1) Manual resizing method:
We provided the surgeonwith pre-designed partial skulls with the generated defectsand corresponding over-sized implants. The surgeon outlinedthe defect contour of each specimen manually based on hisvisual judgment. He then resized the implant with a hand-held cutting tool.
2) Optical tracking method:
The optical tracking methodused a digitizing instrument and an optical tracking system totrace the defect contour (Fig. 5, a). The defect contours werethen cut by the same cutting robot and with same cuttingparameters described above.
B. Tool Center Point (TCP) Calibration
The transformation between the tip of the spindle tooland the robot arm’s end-effector was calibrated using a pivotcalibration [18]. In this method, we hand-guided the roboticarm to different poses, such that the TCP always touchesthe tip of a fixed pin. The accuracy of the TCP calibration,measured by the TCP error, is shown in Fig. 5.
C. Implant Localization
The oversized CCI was secured on the mounting platformwith bolts for the resizing process (Fig. 1, bottom, right). Inorder to obtain the relative position and orientation of theoversized CCI in the robot space (Fig. 2, right), the robotwas hand-guided so that the TCP touched the tip of eachspherical marker separately.The transformation between the local reference frame F re f of the oversized CCI and the robot base F base can be thencalculated based on the known locations of the markers,described in the F re f , and the relative positions from F base to F TCP : Base T TCP = Base T ee · ee T TCP
D. Experimental Details
The integrated system was set up on a computer runningIntel Core i7-6820HQ @ 2.7GHz CPU. The 3D scanner(Artec Spider) collects data at 15 HZ. The KUKKA robot isoperated using online mode via RoboDK . The registrationbetween the 3D-scanned model and the CT model wasimplemented in Meshlab, an open-source software for meshprocessing [19]. The NDI Polaris optical tracking systemoperates at 10 Hz (0.3 mm tracking accuracy) was used inthe comparison experiment. RoboDK is an offline programming and simulation software for indus-trial robots. https://robodk.com/ Fig. 5. Top: a) collecting the defect contour by optical tracking system.b) scanning the defect contour with a handheld 3D scanner. c) robot TCPcalibration errors
IV. R
ESULT
A. Registration
1. Registration by optical tracking systemThree anatomical markers were artificially added to theoriginal CT models and were 3D printed with the defectspecimens. The anatomical points on the printed specimenswere localized in the optical tracking system with a trackinginstrument (Fig. 5 ,a). Then, they were registered back to theCT coordinate system using a singular value decompositionalgorithm [20]. The registration error was given by the meanCartesian distance between the registered anatomical pointsand the original anatomical points in the CT model (TableI).2. Registration by 3D scannerAfter scanning the defect specimen, we first manuallyaligned the three anatomical points on the defect with theoriginal anatomical points defined in the CT model as aninitialization. Then ICP was used to fine-tune the registrationof the scanned specimen to the original CT model. The errorwas evaluated by calculating the mean distance between allof the registered points and their corresponding points in theoriginal CT model (Table I).
B. Evaluation of resized implants
The post-completed implants were fitted to the defectspecimens. They were then 3D scanned and registered tothe original CT models using the anatomical points on thedefect specimens. Each post-completed implant was thenindividually scanned by 3D scanner and registered to theprevious scan of the implant fitted to the defect, so that theimplant can be put to the correct position relative to theground truth defect. ig. 6. An example of gap distance analysis (the first specimen). Left,middle, right show the results of conventional manual modification, 3Dpositioning system, and 3D scanning system, respectively. Top and bottomshow their overviews and their zoomed views. The color bars in the bottomplots show the gap distance between the implant’s boundary and its fitteddefect wall.
We then evaluated the gap distances between the bound-aries of the cut implants and their respective defect walls.The gap distances were visualized in Meshlab [19] (Fig. 6).We used the gap distance distribution (maximum, mean andstandard deviation) to quantify the error for each methodfor the 6 specimens (Fig. 7). Among the mentioned threemethods, our robot-3d-scan integrated method was the onlyone with the mean gap distance below 1.5 mm (Fig. 8).
C. Time cost
Not only does our method achieve the best accuracy, it isalso comparable in the time it takes to utilize the conventionalmanual method (Table II), and it would not affect the ongoingsurgery time.
TABLE IR
EGISTRATION E RRORS OF T WO M ETHODS
S1 S2 S3 S4 S5 S63D positioningsystem (mm) 0.39 0.38 0.32 0.32 0.30 0.343D scanningsystem (mm) 0.04 0.02 0.04 0.04 0.05 0.03
TABLE IIT
IME SPAN OF TWO METHODS ( MINUTES ) Surgeon Proposed methodData acquisition 1 3D Scan 1patient-CT registration 1toolpath generation 1Implant modification 4 - 8 implant localization 1robot execution 2 - 3Total time 5 - 9 6 - 7
V. D
ISCUSSION
We present a novel method for intraoperatively fabricatingprecise CCI in single-stage cranioplasty. In the proposedmethod, we first scan the defect to create a mesh model. Themesh model is then registered to the reconstructed 3D from
Fig. 7. Cutting performance. Visualization of the defect contour (bluecurve) and implant contour (red curve) for a) conventional manual modifi-cation, b) 3D positioning system, and c) 3D scanning system. The numbersin the middle of each plot show the maximum gap distance.Fig. 8. Cutting accuracy evaluation. Mean and standard deviation of thegap distance between the implant and the defect of conventional manualmodification (blue), 3D positioning system (orange), and 3D scanningsystem (green) for six specimens.
CT data in order to define the contour of the defect. Next,a cutting toolpath is generated using the discretized defectcontour. After localizing the oversized CCI in the robot’sbase frame, an automatic cutting process is implemented toenerate the final CCI implant.The proposed method improves the accuracy of the cut by56% compared to the surgeon’s cut and 42% compared to theoptical tracking method. Moreover, the implant cut bound-aries as created by the robot were considerably smoother thanthose created by the expert surgeon. The smooth boundarymay contribute to the better fit of implants to the defect areain actual surgical scenarios. Our proposed method, however,only moderately reduced the operation time compared to theexpert surgeon. Of note, the expert surgeon’s performancetime (5-9 minutes) in this study was significantly belowthe lower spectrum of 10-80 minutes as reported by Berliet al [3]. The robotic modification of oversized CCIs wereshown to be more consistent and accurate when comparedto the expert surgeon’s performance. Of note, We used anavailable seven DOF Kuka robot to perform the cutting tasks.However, a cheaper six DOF robot or a five-axis laser cuttingmachine (e.g. [21]) can also successfully perform smoothcutting as proposed for this research.Some of the limitations of the current study are as follows:1) During the toolpath generation process, the cut angledefining the tool axis attached to the discretized points alongthe defect contour were constant. In actual surgical scenarios,This may cause problems in fitting the implant if the defectboundary is not beveled uniformly. The extension of thiswork will include the development of an algorithm that canextract the bevel angle of the defect wall from the scandata. 2) In this study, the manually-tuned, experimentally-determined cutting speed and spinning rate of the tool werenot optimized. Additional experiments are needed to evaluatethe optimal cutting parameters for smooth cutting of theimplant. 3) In the clinical setting, due to the minimallyexposed surgical area (Figure 1, Top, iii), the process ofregistering a patient’s defect scan to the CT model may bechallenging. A possible remedy is to use two separate 3Dscans at different times during the procedure. Prior to drapingof the patient, marks will be drawn within the surgical area.The first scan will acquire the full exposed head as well as thedrawn marks and register this head model to the preoperativeCT scan of the patient’s skull. After draping the surgical areaand subsequent skull resection, a second scan containingthe defect area information will be registered to the firstscan using the information obtained from the drawn marks.Thereby, the defect scan can be mapped to the CT modelthrough the intermediate scan obtained prior to draping.R
EFERENCES[1] S. Aydin, B. Kucukyuruk, B. Abuzayed, S. Aydin, and G. Z. Sanus,“Cranioplasty: review of materials and techniques,”
Journal of neuro-sciences in rural practice , vol. 2, no. 2, p. 162, 2011.[2] C. M. Pasick, K. Margetis, G. F. Santiago, C. Gordon, and P. J. Taub,“Adult cranioplasty,”
Journal of Craniofacial Surgery , vol. 30, no. 7,pp. 2138–2143, 2019.[3] J. U. Berli, L. Thomaier, S. Zhong, J. Huang, A. Quinones, M. Lim,J. Weingart, H. Brem, and C. R. Gordon, “Immediate single-stagecranioplasty following calvarial resection for benign and malignantskull neoplasms using customized craniofacial implants,”
Journal ofCraniofacial Surgery , vol. 26, no. 5, pp. 1456–1462, 2015. [4] A. M. Shah, H. Jung, and S. Skirboll, “Materials used in cranioplasty:a history and analysis,”
Neurosurgical focus , vol. 36, no. 4, p. E19,2014.[5] A.-R. Fathi, S. Marbacher, and A. Lukes, “Cost-effective patient-specific intraoperative molded cranioplasty,”
Journal of craniofacialsurgery , vol. 19, no. 3, pp. 777–781, 2008.[6] S. Marbacher, L. Andereggen, S. Erhardt, A.-R. Fathi, J. Fandino,A. Raabe, and J. Beck, “Intraoperative template-molded bone flapreconstruction for patient-specific cranioplasty,”
Neurosurgical review ,vol. 35, no. 4, pp. 527–535, 2012.[7] H. Eufinger, A. R. Wittkampf, M. Wehm¨oller, and F. W. Zonneveld,“Single-step fronto-orbital resection and reconstruction with individualresection template and corresponding titanium implant: a new methodof computer-aided surgery,”
Journal of Cranio-Maxillofacial Surgery ,vol. 26, no. 6, pp. 373–378, 1998.[8] G. Gerbino, F. A. Bianchi, E. Zavattero, F. Tartara, D. Garbossa, andA. Ducati, “Single-step resection and reconstruction using patient-specific implants in the treatment of benign cranio-orbital tumors,”
Journal of Oral and Maxillofacial Surgery , vol. 71, no. 11, pp. 1969–1982, 2013.[9] F. Jalbert, S. Boetto, F. Nadon, F. Lauwers, E. Schmidt, and R. Lopez,“One-step primary reconstruction for complex craniofacial resectionwith peek custom-made implants,”
Journal of Cranio-MaxillofacialSurgery , vol. 42, no. 2, pp. 141–148, 2014.[10] P. Dodier, F. Winter, T. Auzinger, G. Mistelbauer, J. M. Frischer,W.-T. Wang, A. Mallouhi, W. Marik, S. Wolfsberger, L. Reissig et al. , “Single-stage bone resection and cranioplastic reconstruction:comparison of a novel software-derived peek workflow with thestandard reconstructive method,”
International journal of oral andmaxillofacial surgery , vol. 49, no. 8, pp. 1007–1015, 2020.[11] A. Wolff, G. F. Santiago, M. Belzberg, C. Huggins, M. Lim, J. Wein-gart, W. Anderson, A. Coon, J. Huang, H. Brem et al. , “Adult cran-ioplasty reconstruction with customized cranial implants: preferredtechnique, timing, and biomaterials,”
Journal of Craniofacial Surgery ,vol. 29, no. 4, pp. 887–894, 2018.[12] R. J. Murphy, K. C. Wolfe, P. C. Liacouras, G. T. Grant, C. R. Gordon,and M. Armand, “Computer-assisted single-stage cranioplasty,” in . IEEE, 2015, pp. 4910–4913.[13] S. Liu, W.-L. Huang, A. Shin, C. Gordon, and M. Armand, “A portableprojection mapping device for medical augmented reality in single-stage cranioplasty,” in
Optical Architectures for Displays and Sensingin Augmented, Virtual, and Mixed Reality (AR, VR, MR) , vol. 11310.International Society for Optics and Photonics, 2020, p. 1131007.[14] A. Haleem and M. Javaid, “3d scanning applications in medical field:a literature-based review,”
Clinical Epidemiology and Global Health ,vol. 7, no. 2, pp. 199–210, 2019.[15] P. J. Besl and N. D. McKay, “Method for registration of 3-d shapes,”in
Sensor fusion IV: control paradigms and data structures , vol. 1611.International Society for Optics and Photonics, 1992, pp. 586–606.[16] C. B. Sullivan and A. A. Kaszynski, “Pyvista: 3d plotting and meshanalysis through a streamlined interface for the visualization toolkit(vtk),”
Journal of Open Source Software , vol. 4, no. 37, p. 1450, 2019.[17] M. Meyer, M. Desbrun, P. Schr¨oder, and A. H. Barr, “Discretedifferential-geometry operators for triangulated 2-manifolds,” in
Vi-sualization and mathematics III . Springer, 2003, pp. 35–57.[18] W. Birkfellner, F. Watzinger, F. Wanschitz, R. Ewers, andH. Bergmann, “Calibration of tracking systems in a surgical envi-ronment,”
IEEE transactions on medical imaging , vol. 17, no. 5, pp.737–742, 1998.[19] P. Cignoni, M. Callieri, M. Corsini, M. Dellepiane, F. Ganovelli, andG. Ranzuglia, “Meshlab: an open-source mesh processing tool.” in
Eurographics Italian chapter conference , vol. 2008. Salerno, 2008,pp. 129–136.[20] K. S. Arun, T. S. Huang, and S. D. Blostein, “Least-squares fittingof two 3-d point sets,”
IEEE Transactions on pattern analysis andmachine intelligence , no. 5, pp. 698–700, 1987.[21] J. Liu, J. Fang, R. J. Murphy, C. Gordon, and M. Armand, “Designand development of 5-axis cranial implant laser cutting system,” in