Back to LuminaBone
🦴 LuminaBone

CavityScan — Technical Protocol v0.1

Protocol proposal May 2026 10 min read

Technical Protocol Proposal — Lumina-Bone Project, Extension Module. CavityScan: photometric stereo 3D mapping of reamed bone cavities for pre-implantation contact prediction.

Protocol IDLB-EXT-001
Version0.1 — Draft
ProjectLumina-Bone (Extension)
DateMay 2026
Duration10–12 weeks
Budget< $800 (parts)
StatusConcept / Pre-prototype
RegulatoryResearch only — no patient use
Research only — no patient use. Regulatory status of protocol LB-EXT-001 (v0.1, draft): concept / pre-prototype. Nothing described here is cleared or intended for clinical or patient use.

1. Background and Clinical Motivation

1.1 The Press-Fit Problem

Cementless orthopedic implants — acetabular cups, femoral stems, tibial baseplates — depend entirely on intimate bone-implant contact to achieve the primary mechanical stability that allows bone ingrowth (osseointegration). The physics is well established: gaps larger than 50 µm at the bone-implant interface prevent bone ingrowth and lead instead to fibrous tissue formation, progressive micromotion, and ultimately aseptic loosening — the leading cause of implant revision surgery.

The reaming step is where this gap problem originates. Surgeons ream the bone cavity to a nominally ideal geometry (hemisphere, taper, or flat tray) to match the implant. In practice, however:

The result: surgeons rely solely on tactile and auditory feedback during impaction to judge fixation quality. This is inherently subjective, non-reproducible, and provides no spatial information about which zones of the interface are in contact and which have gaps.

1.2 The Unmet Need

A review of the competitive landscape confirms that no commercial or research device provides spatial, real-time, intraoperative mapping of the bone cavity geometry before implant placement:

TechnologyWhat It MeasuresBone-Cavity Geometry?Spatial Map?
VERASENSE (OrthoSensor)Medial/lateral joint pressure (soft tissue balance)NoNo
RFA / ISQ (Osstell)Single global implant stability numberNoNo
Impact / acoustic analysisInsertion endpoint signal (global)NoNo
Pressure-sensitive film (Fujifilm Prescale)Contact pressure map — ex vivo onlyIndirectlyYes — lab only
CavityScan (this proposal)Dense 3D map of reamed cavity geometryYESYES

CavityScan directly addresses this gap. By inserting the Lumina-bone photometric stereo endoscope into the reamed cavity after preparation but before implant insertion, a dense 3D surface map of the actual bone cavity geometry is acquired in under 30 seconds. This map is then compared computationally to the nominal implant geometry to produce a predicted contact distribution map — identifying high-contact zones, partial-contact zones, and regions with clinically significant gaps — before any final implant is committed.

1.3 Relationship to Lumina-Bone

CavityScan is a direct extension of the Lumina-bone project. It uses the same hardware (6 mm photometric stereo endoscope with sequential off-axis LEDs), the same core algorithm (near-light Lambertian + inverse-square photometric stereo), and the same software pipeline (Python + OpenCV/PyTorch, Poisson depth integration). The extension contributions are:

2. Project Objectives

The primary objective of CavityScan is to demonstrate that the Lumina-bone photometric stereo pipeline can accurately reconstruct the 3D geometry of a reamed bone cavity from within the cavity, and that this geometry can be used to predict the spatial distribution of bone-implant contact before implant placement.

Specific objectives:

3. Technical Approach

3.1 Physical and Algorithmic Basis

CavityScan inherits the core physical model from Lumina-bone.

Image irradiance model: I(p) = ρ · (n · L̂) / r² where ρ = surface albedo, n = surface normal, L̂ = unit light direction, r = source-to-point distance.

With two or more sequentially illuminated images from known LED positions, the photometric stereo system yields per-pixel surface normals algebraically. Integration of the normal field via weighted Poisson reconstruction recovers the depth map (surface height). Lumina-bone has already validated this pipeline on a synthetic sphere with 2.75% mean depth error — within the range of the in-vivo colonoscopy benchmark (7%, Batlle et al. 2022).

The cavity scanning problem introduces two adaptations relative to the open-surface Lumina-bone application:

3.2 Contact Prediction Module

Once the cavity surface is reconstructed as a point cloud C, the predicted implant-bone contact is computed as follows:

  1. Load the nominal implant surface mesh M (from implant CAD or manufacturer mesh).
  2. Align M to C using the known reaming depth and axis (from navigation system or manual input).
  3. For each point on M, compute the nearest distance d to C. Points with d < threshold τ (default τ = 50 µm, the clinically accepted osseointegration gap limit) are classified as contact zones; points with d > τ are classified as gap zones.
  4. Render a color-coded contact map on the implant surface: green (d < 50 µm, good contact), yellow (50–150 µm, marginal), red (> 150 µm, gap — osseointegration unlikely).

The threshold values are grounded in established biomechanical evidence: micromotion above 30–50 µm causes only partial ingrowth; micromotion above 150 µm completely inhibits bone ingrowth (Szmukler-Moncler et al. 1998; Pilliar et al. 1986). These thresholds are configurable by the user.

3.3 Hardware Configuration

The CavityScan prototype uses the Lumina-bone 6 mm endoscope with the following adaptations:

ComponentSpecification / Adaptation
Camera1.5–2 mm CMOS, same as Lumina-bone baseline; short focal length for near-field focus at 5–20 mm WD
LEDs2× off-axis micro-LEDs at 1.8–2.5 mm radial offset; PWM-sequenced; same as Lumina-bone baseline
Distal tip OD6 mm — fits reaming canals of all standard total hip, knee, and shoulder systems
Shaft length150–200 mm rigid shaft to reach acetabular cavity depth through standard MIS approaches
Scanning motionManual or motorized axial rotation (360°) plus 2–3 axial positions; total of 6–8 endoscope poses per cavity
TrackingPassive optical marker on shaft for pose estimation relative to cavity axis (optional: use navigation system integration)
Total scan timeTarget: < 30 s image acquisition; < 30 s reconstruction and display (< 60 s total)

3.4 Software Architecture

The software pipeline extends Lumina-bone's existing photometric_stereo_sphere.py with three additional modules:

ModuleFunctionTechnology
cavity_scan.pyOrchestrates multi-pose image acquisition; triggers LED sequencing; stores pose metadataPython + OpenCV
cavity_stitch.pyRegisters partial depth maps from each endoscope pose into a unified cavity point cloud using ICPOpen3D / PCL
contact_predict.pyLoads implant CAD mesh; aligns to reconstructed cavity; computes per-point gap distances; outputs color-coded contact maptrimesh / NumPy
cavity_display.pyReal-time visualization of contact map overlaid on implant geometry; surgeon-facing GUI with traffic-light color codingOpen3D / PyQt

4. Validation Protocol

4.1 Phantom Preparation

4.1.1 Cavity Phantoms

Three cavity geometry types will be fabricated, corresponding to the most common cementless implant interfaces:

Ground-truth cavity geometry for each phantom will be acquired by: (a) high-precision laser scanning (Faro Focus or equivalent, accuracy ≤ 0.1 mm) before and after intentional defect machining, and (b) micro-CT at 100 µm resolution for the foam phantoms.

4.1.2 Implant Reference Models

Nominal implant CAD meshes will be obtained from open-source orthopedic implant libraries (e.g., OrthoLoad, manufacturer-provided meshes for standard implant sizes used in the phantoms). One acetabular cup (48 mm), one tibial baseplate (size M), and one femoral stem (size 3) will serve as reference models.

4.2 Data Acquisition Protocol

For each cavity phantom:

  1. Insert CavityScan endoscope to defined depth markers on the shaft.
  2. Acquire 2-LED sequential images at 0°, 60°, 120°, 180°, 240°, 300° rotation (6 angular positions per axial depth).
  3. Repeat at 2–3 axial depth positions within the cavity (proximal, mid, distal zones).
  4. Record endoscope pose at each position (optical tracker or fiducial-based).
  5. Process: run photometric stereo per frame → stitch → compute contact map.
  6. Repeat acquisition 5 times per phantom for intra-session repeatability assessment.

Three operators will each perform the acquisition independently on the same phantom to assess inter-operator variability.

4.3 Primary Outcome Metrics

MetricDefinitionAcceptance Criterion
Cavity reconstruction accuracy — MAEMean absolute depth error vs. laser scan ground truthMAE < 200 µm
Cavity reconstruction accuracy — RMSERoot mean square depth error vs. ground truthRMSE < 300 µm
Defect detection rate% of intentional defects (100, 200, 300 µm ridges) correctly identified in contact map≥ 90% for ≥ 200 µm defects
Contact zone agreementDice coefficient between CavityScan predicted contact map and ground-truth contact map (from laser scan + implant mesh overlap)Dice ≥ 0.80
Intra-session repeatabilityStandard deviation of MAE across 5 repeated acquisitions of same phantomσ < 100 µm
Inter-operator variabilityRange of MAE across 3 operatorsRange < 150 µm
Total workflow timeTime from endoscope insertion to contact map displayed on screen< 60 seconds

4.4 Secondary Analyses

5. Project Timeline

PhaseWeeksActivities
Phase 11–2System design & procurement: cavity scanning geometry analysis; shaft length specification; phantom machining plan; parts ordering (bone substitute blocks, optical tracker markers, shaft hardware)
Phase 23–4Phantom fabrication & ground truth: machine hemispherical cavities with intentional defects; laser scan all phantoms; acquire micro-CT; establish ground-truth contact maps for each phantom + implant combination
Phase 35–6Cavity scanning adaptation: adapt Lumina-bone endoscope shaft for cavity insertion; develop cavity_scan.py pose-acquisition loop; calibrate LED positions for enclosed-cavity geometry; validate photometric stereo on single-pose cavity images
Phase 47–8Multi-view stitching & contact prediction: implement cavity_stitch.py ICP registration; implement contact_predict.py gap computation; develop color-coded contact map visualization; end-to-end pipeline integration test
Phase 59–10Validation experiments: systematic phantom testing per Section 4 protocol; 3-operator inter-variability study; timing benchmarks; quantitative metric computation vs. ground truth
Phase 611–12Analysis, reporting & dissemination: statistical analysis; ablation studies; figures; technical report; demo video; open-source code release (GitHub); conference abstract submission

6. Resources and Budget

ItemEst. Cost (USD)Notes
Polyurethane foam bone substitute blocks (Sawbones)$12020 PCF + 40 PCF; ×3 geometry types
Composite bone femur phantom (Sawbones 4th gen)$180For femoral stem canal testing
CNC machining of intentional defects in phantoms$80In-house or machine shop
Optical tracker markers / fiducials$60Passive reflective markers
Extended shaft hardware (aluminium tube, connectors)$50150–200 mm rigid shaft
3D printing materials (ABS/PETG for tip adapters)$30Cavity entry guides, tip jigs
Implant CAD meshes (open-source / manufacturer)$0OrthoLoad library / free DXF files
Miscellaneous (cables, adhesives, spare LEDs)$50Consumables
TOTAL (parts only)~$570GPU workstation & laser scanner: existing lab assets

Existing Lumina-bone assets reused without additional cost: 6 mm endoscope prototype; PWM LED driver electronics; GPU workstation; Python + OpenCV/PyTorch software stack; photometric stereo pipeline (photometric_stereo_sphere.py and associated modules).

7. Expected Deliverables

8. Risks and Mitigations

RiskLikelihoodMitigation
Insufficient photometric stereo accuracy at short WD inside curved cavity (reflection geometry differs from open-surface)MediumRe-calibrate LED positions for inside-cavity geometry; add 3rd LED if 2-light solution is underdetermined; fall back to SFS if required
Specular reflections from damp bone surface degrading normal estimationMediumApply existing Lumina-bone specular masking pre-step; test on dry vs. irrigated phantoms; use bone phantoms coated with matte sealant for initial validation
ICP stitching fails due to insufficient overlap between adjacent endoscope posesLow–MediumDesign acquisition protocol with ≥ 30% overlap between adjacent poses; use cavity axis as common reference for initial pose estimate before ICP
Reconstruction accuracy insufficient for 100 µm defect detectionMediumAcceptance criterion scoped to ≥ 200 µm defects (clinical threshold for osseointegration); 100 µm remains a stretch goal
Workflow time exceeds 60-second targetLowGPU acceleration of Poisson solver (cuSPARSE); reduce pose count to 4 if accuracy permits; pre-load implant mesh before scan
Implant CAD mesh not available for target sizesLowGenerate synthetic meshes from measured implant geometry using structured light scanner (available in lab)

9. Future Directions

Successful completion of this protocol establishes the technical foundation for the following downstream developments:

10. References

  1. C. Wu, S. G. Narasimhan, and B. Jaramaz, "A multi-image shape-from-shading framework for near-lighting perspective endoscopes," Int. J. Comput. Vis., vol. 86, no. 2-3, pp. 211-228, 2010, doi: 10.1007/s11263-009-0207-3.
  2. V. Parot, D. Lim, G. Gonzalez, N. S. Nishioka, B. J. Vakoc, and N. J. Durr, "Photometric stereo endoscopy," J. Biomed. Opt., vol. 18, no. 7, p. 076017, 2013, doi: 10.1117/1.JBO.18.7.076017.
  3. V. M. Batlle, J. M. M. Montiel, and J. D. Tardos, "Photometric single-view dense 3D reconstruction in endoscopy," in Proc. IEEE/RSJ Int. Conf. Intelligent Robots Systems (IROS), 2022, doi: 10.1109/IROS47612.2022.9981742.
  4. R. M. Pilliar, J. M. Lee, and C. Maniatopoulos, "Observations on the effect of movement on bone ingrowth into porous-surfaced implants," Clin. Orthop. Relat. Res., vol. 208, pp. 108-113, 1986.
  5. S. Szmukler-Moncler, H. Salama, Y. Reingewirtz, and J. H. Dubruille, "Timing of loading and effect of micromotion on bone-dental implant interface: review of experimental literature," J. Biomed. Mater. Res., vol. 43, no. 2, pp. 192-203, 1998.
  6. MedicalTek, "MonoStereo 3D endoscopic imaging system (MS-301/MS-302)," MedicalTek Co., Ltd., Taiwan, 2024-2025 [Online]. Available: medicaltek.biz.
  7. W. Yuan, S. Dong, and E. H. Adelson, "GelSight: High-resolution robot tactile sensors for estimating geometry and force," Sensors, vol. 17, no. 12, p. 2762, 2017, doi: 10.3390/s17122762.
  8. T. L. Bobrow et al., "Multi-contrast laser endoscopy for in vivo gastrointestinal imaging," npj Imaging, 2025, doi: 10.1038/s44303-026-00161-y.
  9. N. J. Durr et al., "System for clinical photometric stereo endoscopy," Proc. SPIE 8935, Advanced Biomedical and Clinical Diagnostic Systems XII, 89351F, 2014.
  10. "Quantitative assessment of prosthesis press-fit fixation," U.S. Patent 11,974,876, 2024.