Laser scanning registration for knee arthroplasty Authors: Thomas Warren, Ulani Hayter Otaola, Brett Robertson, Riaz Khan, Sam Withers, Hari Goonatillake, Benjamin Hewitt, Tomasz Kozak, Sajan Kuriakose, Tareq Batanony, Jon Greenwald, Morteza Meftah Published: 2026 ABSTRACT AbstractAimsAccurate registration of anatomy is fundamental to robotic and computer-assisted surgery; however, conventional methods rely on manual data acquisition performed by the surgeon using a tracked probe that is slow and affected by operator variability. Limitations of manual methods have hindered wider adoption of surgical navigation despite its benefits for alignment accuracy, particularly in orthopaedic applications. This study evaluates a laser surface scanning system, LumaScan, against standard intraoperative probe-based registration with a controlled phantom knee model.MethodsSurgeons performed repeated probe-based data acquisition for registration, while LumaScan executed operator-initiated scans mounted to a robotic arm. The accuracy of each modality was assessed using root mean square (RMS) error relative to a CT-derived reference model, alongside evaluations of repeatability and acquisition time.ResultsLumaScan achieved a RMS error of 0.07 mm (SD 0.02), compared to probe-based methods, which resulted in a RMS error of 0.34 mm (SD 0.36). Acquisition time was reduced from 112 seconds (SD 75.8) manually to 6.38 seconds (SD 0.03) for LumaScan (p < 0.001), with probe-based timing increasing three times from the fastest to the slowest participant despite a fixed protocol. The errors of the femur and tibia achieved by LumaScan had a > 70% reduction compared to manual surface acquisition, with these results being statistically significant (p < 0.001). The results demonstrate that robotic laser scanning can significantly improve speed, accuracy, and precision over manual methods. Minimizing operator influence and standardizing registration has the potential to reduce error sources that persist in clinical workflows, subsequently accelerating the integration of navigation and robotics into routine orthopaedic procedures.ConclusionLumaScan enables faster, more accurate, and more precise anatomical registration than manual methods.Cite this article: Bone Joint Res 2026;15(9):1092–1100.Keywords: Laser scanning, Registration, Knee, Arthroplasty, Robotic, knee, knee arthroplasty, root mean square, femur, tibia, robotic arm, orthopaedic procedures, total knee arthroplasty (TKA), arthroplasty, orthopaedic surgeons Article focus Evaluate whether a robotic laser surface scanner (LumaScan) yields lower registration error than conventional probe-based methods in image-guided total knee arthroplasty (TKA). Evaluate whether a robotic laser surface scanner (LumaScan) yields lower registration error than conventional probe-based methods in image-guided total knee arthroplasty (TKA). Test whether LumaScan reduces acquisition time and operator-dependent variability in a simulated clinical knee model. Test whether LumaScan reduces acquisition time and operator-dependent variability in a simulated clinical knee model. Quantify performance using root mean square (RMS) error vs CT-derived reference, repeatability (σ), and acquisition time. Quantify performance using root mean square (RMS) error vs CT-derived reference, repeatability (σ), and acquisition time. Key messages LumaScan delivered sub-mm accuracy (RMS error ≈ 0.07 mm) with ~ 80% lower error than manual probe mapping (≈ 0.34 mm; p < 0.001). LumaScan delivered sub-mm accuracy (RMS error ≈ 0.07 mm) with ~ 80% lower error than manual probe mapping (≈ 0.34 mm; p < 0.001). ~17 × faster capture (≈ 6 to 7 s vs ≈ 112 s) with minimal spread, indicating improved reproducibility and reduced operator influence. ~17 × faster capture (≈ 6 to 7 s vs ≈ 112 s) with minimal spread, indicating improved reproducibility and reduced operator influence. Dense, contactless point clouds can standardize registration, addressing known adoption barriers for computer-assisted surgery/robotic TKA (speed and consistency) without sacrificing accuracy. Dense, contactless point clouds can standardize registration, addressing known adoption barriers for computer-assisted surgery/robotic TKA (speed and consistency) without sacrificing accuracy. Strengths and limitations This is a controlled, head-to-head comparison on the same phantom with identical tracking and iterative closest point pipeline; high sampling density; clear timing; and accuracy endpoints. This is a controlled, head-to-head comparison on the same phantom with identical tracking and iterative closest point pipeline; high sampling density; clear timing; and accuracy endpoints. This study used a phantom model without soft-tissue, bleeding, or motion, and used a single manual comparator workflow. This study used a phantom model without soft-tissue, bleeding, or motion, and used a single manual comparator workflow. The results warrant cadaveric and in vivo validation to confirm performance under soft-tissue, fluid, and motion artifacts and across alternative navigation platforms. The results warrant cadaveric and in vivo validation to confirm performance under soft-tissue, fluid, and motion artifacts and across alternative navigation platforms. Introduction Ensuring accurate alignment during the process of total knee arthroplasty (TKA) is fundamental to positive clinical outcomes, including implant longevity and functional restoration.1 Deviations in alignment are associated with increased wear, loosening, and revision rates, with studies reporting increased risk of failure when coronal misalignment exceeds 3°.2 Computer-assisted surgery (CAS) systems have become a common integration with navigation and robotic technologies, improving consistency of component positioning, allowing for patient specific surgical plans, and reducing alignment outliers compared with conventional instrumentation.3-6 However, the potential benefits of CAS depend on the fidelity and efficiency of intraoperative registration, which remains a manual bottleneck in many workflows. In image-based CAS, the patient’s anatomy is intraoperatively registered to a preoperative image-derived 3D model (typically CT or MRI), generating a best-fit rigid transformation between physical space and model space for navigation or robot execution.7 Registration may be performed using fiducial-based methods and/or surface matching, which requires manual intraoperative point acquisition with a tracked probe touching anatomical landmarks on the bone and cartilage surface. In imageless CAS, no preoperative image model is used; instead, the anatomical axes and reference frames are determined from digitized landmarks and sparse surface points acquired intraoperatively with the probe.7 In both workflows, the quality of the registration is influenced by the number and distribution of collected points, subjective landmark identification, and the available surgical exposure, introducing operator-dependent variability and limiting repeatability.8,9 These limitations are clinically and operationally relevant. Registration error can spread to component positioning and contribute to alignment outliers, particularly in regions of restricted exposure or variable morphology where consistent point access is challenging.2,9 Significantly, manual point acquisition is also time-consuming, and contributes to the learning curve and intraoperative workflow disruption that can restrict broader adoption of navigation and robotics despite alignment benefits.9-11 As a result, there is an interest in acquisition approaches that reduce operator dependency and time spent while maintaining registration accuracy. Laser surface scanning offers a potential solution by rapidly capturing dense, high-resolution surface geometry (point clouds) without contact. Although early feasibility studies have examined laser scanning in orthopaedic applications, its use in arthroplasty remains underexplored, with limited direct evaluation against probe-based registration within image-guided TKA workflows.9,12-15 LumaScan (OrthoLuma, Australia) is a 3D laser scanner designed to address the limitations of manual registration by acquiring high-density surface data in seconds. This study evaluates LumaScan in comparison to standard intraoperative probe-based registration in a controlled knee phantom setting, assessing registration accuracy, repeatability (precision), and acquisition time against a CT-derived reference model. The hypothesis of this study is that LumaScan will demonstrate a lower root mean square (RMS) error relative to manual methods in a simulated clinical setting, with improved precision and reduced acquisition time. Methods This study assessed the performance of LumaScan, a laser surface scanner, as an alternative to conventional probe-based methods in a simulated clinical environment. A total of 60 manual registration acquisitions by six surgical professionals (RK, HG, BH, TK, SK, TB) (n = 10 per participant) and six LumaScan acquisitions were performed. All iterations were performed on the same model. Accuracy was defined as the RMS error between each acquired point cloud and a high-resolution anatomical reference model. Precision was determined by analyzing the standard deviation of RMS error across repeated trials, and acquisition time was recorded to assess workflow efficiency and operator consistency. All iterations were performed with the aforementioned pipeline, outlined in Figure 1, and no data were lost or excluded. Experimental workflow for comparing LumaScan (Australia) laser surface acquisition with conventional probe-based registration in total knee arthroplasty. RMS, root mean square. System and equipment The model used throughout the study was a standardized knee joint phantom, in which the femur and tibia were 3D-printed in polylactic acid derived from a CT scan of an anonymous patient. The CT dataset was provided by CTEC, Medical School, University of Western Australia, and was anonymized prior to investigator access. The ethics/governance requirements for secondary use of this dataset were obtained in accordance with institutional policy. A fellowship-trained orthopaedic surgeon (RK) prepared the model by mounting the knee phantom in a supine position at 90° flexion. A disposable surgical drape was then positioned over the phantom to occlude the surrounding area and approximate typical intraoperative exposure (i.e., restricted access due to soft-tissue and skin coverage). A 10 cm medial parapatellar incision was created to replicate the standard surgical approach. As shown in Figure 2, this setup emulates the constrained visual and physical corridor encountered intraoperatively during a standard parapatellar exposure. Passive infrared fiducial markers (Polaris Vega, NDI, Canada) were anchored to the bone by the same surgeon, positioned into the anteromedial distal femoral metaphysis and the proximal metaphyseal region.16 The phantom and scanning setup were fixed in position throughout data collection in order to prevent movement-related variability. Simulated clinical phantom model with fiducial markers and draping as used in the study. Probe-based acquisition A convenience sample of six participants was recruited, which consisted of four orthopaedic surgeons, one orthopaedic registrar, and one senior robotics nurse (RK, HG, BH, TK, SK, TB) known to the investigators and available during the study timeframe. Power calculations (α = 0.05, power = 0.8) for the primary endpoint (RMS surface distance, mm) were used as a planning guide, assuming a minimum detectable difference of 0.2 mm and a SD of 0.32 mm. In the absence of directly comparable variance estimates, this SD was conservatively proxied from Vendries et al,17 who reported CT-versus-structured-light-scanner RMS surface distances of 0.95 mm (SD 0.32) (n = 3) following surface matching. Given the exploratory nature of this study and the intention to characterize operator-dependent variability in the manual workflow, we oversampled probe-based acquisitions (n = 60) and collected six LumaScan acquisitions (n = 6). Each study participant performed five trials using manual technique on two separate sessions, in which the tool was swept over the joint surface in multiple regions of the exposed articular surface and adjacent bony contours to generate a point cloud. Acquisitions were performed on the femur first and then the tibia. For this study a passive, four marker, blunt tip probe was provided and used to lightly track across the exposed bone surfaces. Surgeons were directed to spend three to five seconds on each region, as per a predefined protocol. The regions composed of the medial and lateral femoral condyles (distal articular surface and posterior condyles), the anterior femoral cortex, the intercondylar notch, the medial and lateral epicondyles, the medial and lateral tibial plateau, the anterior tibial tuberosity, and the bony margin of the proximal tibia. Each trial was timed, and if a captured point was interrupted or technical issues displayed, the point was recaptured. LumaScan acquisition LumaScan operates by projecting a laser across a surface (Figure 3) and capturing the reflection to generate sequential cross-sectional datasets, referred to as profiles. Each profile comprises a dense array of spatially resolved data points that distinguish the surface contours in a specific plane. By lofting the sequential sections, a dense and continuous 3D point cloud of the surface is generated. Representation of LumaScan (Australia) laser trajectory. The device exhibits accuracy of 12 µm line linearity across its 100 mm vertical measuring range, supporting high-fidelity surface sampling. It was mounted onto a UR30 collaborative robotic arm, which provides a pose repeatability of ± 0.1 mm, ensuring repeatable and precise execution (Universal Robots, Denmark).18 The UR30 was used to standardize scan execution in this study; the LumaScan is a standalone acquisition module and is not inherently coupled to a specific actuation platform. Each LumaScan acquisition was defined as a single robotic sweep of the target region, producing one point cloud. Six acquisitions were collected, each using the same programmed sweep path. Acquisition time was reported as the effective capture time of the retained region of interest used for registration (derived from scan speed and retained sweep length). The operating parameters were selected to accommodate a range of different surface reflectivity, and the system has been bench-tested during development on representative reflective surgical hardware (including fiducial-marker components) to confirm stable surface capture under typical conditions. Post-processing Post-processing (including region-of-interest handling/artifact management and point-cloud-to-model registration) was performed offline using an internally developed pipeline designed for standardized, objective comparison between LumaScan and conventional probe-based workflows. A CT-derived anatomical surface mesh (generated as part of the existing phantom fabrication pipeline, prior to this study) was used as the common geometrical reference for evaluation. Both the probe-based point set and each LumaScan acquisition (3D point set/point cloud) were independently rigidly registered to this reference surface, and performance was quantified as the RMS surface distance after registration. The processing pipeline was intended for controlled evaluation rather than computational efficiency; therefore, the reported acquisition time reflects surface capture only, and excludes post-processing runtime. These steps are amenable to workflow-specific optimization and, in clinical integration, would be automated and embedded within navigation software to minimize latency and surgeon interaction. Registration within the processing was performed using a two-stage iterative closest point (ICP) procedure.19 A coarse ICP fit was first applied to achieve global rigid alignment by minimizing the mean squared distance between the acquired surface and the reference model. A fine ICP fit was then applied to further reduce point-to-model distances until convergence. The final fitment error (registration accuracy) was calculated as the RMS error of the Euclidean distances between corresponding points in the aligned dataset and the reference model. This is shown in Equation 1, where ||pi-qi|| is the Euclidean distance between each matched point pair and N is the number of matched pairs. Equation 1: RMS error (mm) of the Euclidean distances: Manual acquisitions were performed under the assumption that only intended surface points were successfully captured, the coarse ICP fitment was fair for all cases, and the simulated environment and phantom were representative of the bony anatomy. Compute requirements were modest on standard hardware; workflow time is expected to reduce further with integration and optimization. Statistical analysis Registration accuracy was quantified as RMS surface distance error (mm) for femur and tibia. The differences in registration accuracy were assessed using Welch’s t-test (Equation 2) due to unequal sample sizes and spread between groups. Differences in acquisition time were assessed using the Mann-Whitney U test, given non-normality of the manual time distribution. Statistical significance was defined as p < 0.05. All statistical analyses were performed in Python using the SciPy and NumPy libraries. Equation 2: Welch’s t-test equation: Results The RMS error (µ) and SD (σ) across the trials for both methods are shown in Table I, where the errors of the femur and tibia achieved by LumaScan had a > 70% reduction compared to manual surface acquisition. These results were statistically significant (p < 0.001, Welch’s t-test). Table I.Root mean square error data for both methods reported in mm.AnatomyProbe-based (µ ± σ), mmLumaScan (µ ± σ), mmFemur0.359 (0.348)0.066 (0.002)Tibia0.310 (0.377)0.078 (0.002) Root mean square error data for both methods reported in mm. Probe-based acquisitions required a mean of 112 seconds (SD 76), with time increasing three times from the fastest to the slowest participant, despite a fixed protocol. The speed of each LumaScan acquisition was 14.4 mm/s, resulting in a mean acquisition of 6.38 seconds (SD 0.03). The variability of the LumaScan arose from ROI segmentation/selection affecting the retained sweep length (and in turn the reported effective acquisition time). The difference in acquisition time was statistically significant (Mann-Whitney U = 360, p < 0.001), with the manual distribution right-skewed (median 80 seconds (IQR 63 to 115)). Discussion This study evaluated whether contactless laser surface acquisition can improve intraoperative registration data capture for arthroplasty, relative to the standard approach of surgeon-performed probe registration. Under controlled phantom conditions with identical tracking infrastructure and a common processing pipeline, LumaScan demonstrated significantly lower RMS surface distance error and reduced variability compared with probe-based acquisition, while reducing acquisition time by approximately an order of magnitude. These results support the technical feasibility of dense laser surface capture as a arthroplasty for sparse manual point acquisition during registration in navigated and robotic TKA workflows. Direct comparators in arthroplasty remain limited. In this study, the RMS errors achieved by LumaScan were lower than those reported for intraoperative laser scanners in cadaveric and in vitro models. These typically range from 0.8 mm to 1.3 mm across craniofacial, fresh-frozen knee, MRI-registered cadaver and distal humerus studies.12,15,20-22 Beyond laser scanning, other intraoperative surface acquisition methods have been explored, including structured light scanners, handheld depth cameras, and other optical surface imaging approaches.22-24 These methods can reduce operator dependence relative to probe-based digitization, but reported constraints include line-of-sight occlusion, sensitivity to motion, variable device performance, and added workflow burden from setup, processing, or calibration requirements. The results presented in this study derive from an idealized scenario, a rigid synthetic phantom. In cadaveric and clinical environments, additional sources of error are introduced, including variability in exposure, tissue compliance, instrument occlusion, and the fidelity of the preoperative model to the intraoperative anatomy.25 The statistically significant RMS values (~ 0.07 mm) which we observed demonstrate that, when factors are controlled, dense contactless acquisition can outperform manual digitization in both accuracy and repeatability.15,26,27 In clinical practice, component position is typically evaluated in angular and translational terms (such as coronal/sagittal alignment, femoral rotation, joint line position), and downstream outcomes such as pain, function, survivorship, and revision risk are multifactorial, including influences from implant design and functional biomechanics.1,2,28,29 Recent studies suggest that alignment strategy can be associated with differences in patient-reported outcomes following TKA.30-32 While small surface registration errors can propagate into clinically relevant translational and rotational deviations depending on the registration geometry and landmark distribution, a direct link between an improvement in RMS surface distance and clinical outcomes is not established in this study.28,33 Registration quality is an enabling step for navigation and robotic execution, and improvements in registration repeatability may reduce operator-dependent variability and the risk of registration-driven outliers. In this study, probe-based acquisition demonstrated large dispersion in RMS error, which is consistent with known variability in landmark identification and point acquisition techniques.9,34 Dense surface acquisition, as seen in LumaScan, provides redundancy and reduces sensitivity to point placement and sparse sampling effects, which are known to influence rigid registration error behaviour.35 The mean RMS difference may not be determining, however, the substantial reduction in variability may be clinically meaningful as a pathway to more standardized execution, particularly in workflows where registration remains a manual bottleneck. Time and workflow burden are recognized determinants of CAS adoption.36 In this study, LumaScan consistently completed surface capture in a mean of 6.4 seconds (SD 0.03), compared to manual probe-based capture which required a mean of 112 seconds (SD 75.8), making LumaScan approximately 17 times faster. The spread of manual time was substantial, with participants differing by approximately threefold despite a fixed protocol, once again highlighting operator dependence of manual digitization. This comparison reflects surface capture time only; in this study, post-processing and registration were conducted using a research-level pipeline and are not presented as an optimized end-to-end intraoperative workflow time. Conventional CAS workflows can require five to 11 minutes for manual point cloud acquisition in theatre, contributing to reluctance in adoption.33,37,38 From an operational perspective, reducing the time and variability of registration minimizes potential for workflow disruption, reduces training burden, and offers a predictable, reproducible step that can be integrated into time-critical surgery. Many image-guided TKA workflows use preoperative CT/MRI for planning and to define the reference model for registration.7 The presented work does not evaluate planning accuracy or imaging modality choice; it isolates the intraoperative registration acquisition step by comparing tracked probe digitization with contactless surface capture against a CT-derived reference surface. Workflow integration with clinical CT/MRI planning pathways will be assessed in future studies. In imageless navigation workflow, the intraoperative scanning can provide an accurate, real-time surface morphology that represents accuracy of MRI with cartilage mapping. The primary limitation observed in this study is the use of a synthetic phantom without soft-tissue, bleeding, or motion. These factors may introduce additional capture artifacts for optical scanning and may also exacerbate variability in manual probe contact. Another limitation is that only a single manual probe-based registration modality was evaluated as the comparator; this allowed for standardization of the study but did not account for potential performance differences of other navigation platforms. Additionally, the current implementation used an internal research processing pipeline and does not represent optimized clinical software integration. It is also noted that the current study evaluates registration of the distal femur and proximal tibia only, and does not address acquisition of the hip centre or ankle malleoli required for full mechanical axis determination in computer-assisted TKA. The combined surface-acquisition workflow and phantom design of this study do not support per-landmark accuracy comparisons or direct mapping of RMSE values onto thresholds for clinically meaningful improvement; cadaveric or in vivo evaluation linking registration accuracy to component positioning and patient outcomes would be required to address both. While these limitations are noted, they materially constrain the transferability of the present findings to live surgical use; the accuracy and time gains observed here should be interpreted as evidence of feasibility under controlled conditions rather than as predictive of intraoperative performance, which remains to be established. Future work should progress with the following considerations: 1) cadaveric validation under representative exposure, contamination, and occlusion conditions; 2) quantification of robustness to reflective hardware, fluid films, and variable incidence angles; 3) end-to-end workflow timing including automated processing; and 4) outcome-relevant validation endpoints in clinical series, such as alignment outliers, repeatability of registration across users, and downstream component positioning metrics, before linking technical improvements to longer-term patient outcomes. In conclusion, LumaScan achieved high-precision registrations with < 0.1 mm RMS error for both the distal femur and proximal tibia in less than seven seconds. Furthermore, the device can reduce errors and improve repeatability compared to conventional probe-based registration. In this controlled study, these improvements address recognized limitations of manual surface acquisition, in particular, variability between operators and longer acquisition times, which are barriers to the wider adoption of CAS in TKA.39 While further validation in cadaveric and in vivo settings is required, the results highlight the capacity of LumaScan to deliver point clouds that can be used for surgical registration. Integrating LumaScan into navigation or robotic arthroplasty platforms may enable faster and more consistent registration without compromising accuracy. This could contribute to reducing variability in registration, supporting more standardized component placement, and facilitating broader adoption of CAS workflows. Author contributions T. Warren: Data curation, Formal analysis, Investigation, Methodology, Project administration, Software, Visualization, Writing – original draft, Writing – review & editing U. Hayter Otaola: Data curation, Formal analysis, Investigation, Visualization, Writing – original draft, Writing – review & editing B. Robertson: Conceptualization, Methodology, Project administration, Supervision, Validation, Writing – review & editing R. Khan: Conceptualization, Data curation, Investigation, Methodology, Writing – review & editing S. Withers: Software, Validation, Writing – review & editing H. Goonatillake: Data curation, Investigation, Resources, Writing – review & editing B. Hewitt: Data curation, Investigation, Resources, Writing – review & editing T. Kozak: Data curation, Investigation, Resources, Writing – review & editing S. Kuriakose: Data curation, Investigation, Resources, Writing – review & editing T. Batanony: Data curation, Investigation, Resources, Writing – review & editing J. Greenwald: Conceptualization, Methodology, Writing – review & editing M. Meftah: Conceptualization, Methodology, Validation, Writing – review & editing Funding statement The authors received no financial or material support for the research, authorship, and publication of this article. ICMJE COI statement R. Khan and B. Robertson report support for this study from the Australian Institute of Robotic Orthopaedics and OrthoLuma, as well as funding from the Australian Government which is unrelated to this study. M. Meftah reports stock or stock options in Caira, Biobot, and OrthoLuma, unrelated to this study. U. H. Otaola, B. Hewitt, R. Khan, T. Kozak, M. Meftah, B. Robertson, T. Warren, and S. Withers are affiliated with Arthrolase or Caira Surgical in roles including executive leadership and engineering staff. This represents a potential competing interest due to the companies’ development of related technologies. No additional competing interests are declared. Data sharing The data that support the findings for this study are available to other researchers from the corresponding author upon reasonable request. Ethical review statement This study did not involve human or animal subjects. All experimental work was performed on a synthetic phantom knee model 3D-printed from a fully anonymized CT dataset, provided by CTEC, Medical School, University of Western Australia, with secondary-use approval obtained in accordance with institutional policy. As no living human or animal subjects were involved, formal HREC approval was not required. Open access funding Open access funding was provided by the Australian Institute of Robotic Orthopaedics Pty Ltd (trading as ArthroLase). © 2026 Warren et al. This article is distributed under the terms of the Creative Commons Attribution No Derivatives (CC BY-ND 4.0) licence (https://creativecommons.org/licenses/by-nd/4.0/), which permits the reuse of the work for any purpose, including commercially, provided the original author and source are credited; however, it cannot be distributed to others in any adapted form. Data Availability The data that support the findings for this study are available to other researchers from the corresponding author upon reasonable request. References Associated Data Data Availability Statement The data that support the findings for this study are available to other researchers from the corresponding author upon reasonable request.