How It Works — Augmene

01 Capture Quality Reconstruction accuracy depends on the visual texture of the surface being captured.

Matte, lightly textured surfaces work best. Multi-view reconstruction relies on identifying the same surface points across many frames; uniformly shiny or featureless surfaces provide nothing for the algorithm to match.
A light dusting of baby powder on smooth or reflective areas dramatically improves reconstructability without leaving residue on clinical equipment. Apply sparingly — heavy coating obscures shape detail.
Avoid specular highlights and transparency. Wet skin, silicones, and polished liners reflect differently in each frame and confuse feature matching.
Patient stillness during capture is essential. Any movement of the limb invalidates assumptions that all frames see the same rigid geometry.

If reconstruction shows visibly sparse or noisy regions, the most common cause is a shiny or untextured patch in that part of the limb. A retake with a light powder pass typically resolves it.

02 Lighting Use diffuse, even room lighting and avoid strong shadows or specular highlights.

Prefer indoor overhead lighting or two soft light sources on either side of the patient. Diffuse light produces consistent surface brightness across all viewpoints.
Avoid direct sunlight and single bright lamps. They create moving specular highlights that appear in different pixel positions as the phone orbits.
Avoid deep shadows. Dark regions lose texture and fail to reconstruct; the resulting mesh will have holes under the limb.
Keep the room's lighting stable for the full ~30 second capture — do not pass in front of a window or flip a light switch mid-scan.

03 Capture Method A slow, steady orbit with full coverage gives the best reconstruction.

Orbit the limb at a constant distance, roughly 25–30 cm, keeping the full limb framed and centred throughout the sweep.
Move slowly and steadily — roughly one full orbit in 20–30 seconds. Rapid motion produces motion blur which the reconstruction pipeline cannot recover from.
Aim for > 270° of coverage. Missing sides produce hollow or collapsed meshes on the unseen faces.
Keep the phone roughly perpendicular to the limb axis and vary the vertical angle slightly between low, eye-level, and high during the orbit to capture top and bottom surfaces.
Short video-style capture preferred over burst photos; the in-app capture timing is tuned for this.

04 Calibration & Scale A ChArUco board in view gives the system an authoritative metric reference.

ChArUco board: the printed 5×7 DICT_5X5_1000 board supplied in the app provides known physical dimensions (24 mm squares, 18 mm markers). When the board is visible in the captured frames, the pipeline recovers real-world scale directly from its geometry.
Place the board next to the limb, flat on the same surface, visible in roughly a third of the capture orbit. Partial visibility is fine — the system only needs the board detected in a handful of well-distributed frames.
Fallback when the board is not visible: scale is estimated from the trajectory spread of the phone during capture. This produces a useful but less precise dimension estimate.
Scale source shown on every scan. The clinic scan detail page surfaces whether ChArUco was used and a confidence value so the clinician can weight the reported dimensions accordingly.

ChArUco detectedboard was found in sufficient frames

Scale sourcecharuco, charuco-assisted, or camera-spread

Confidence0–100 %, higher when more inlier pose pairs agree

05 Accuracy & Limitations Research-stage system. Reported dimensions are indicative, not clinically validated.

Augmene is currently in prototype development. The system has not been clinically validated and is not regulatory-cleared for diagnostic use in any jurisdiction.
Dimensional accuracy depends on capture conditions — good lighting, matte surface, visible ChArUco board, and a complete orbit. Under favourable conditions the system produces geometrically consistent shape information; accuracy degrades predictably when conditions drop.
Reported dimensions are indicative, intended to support clinical assessment, not substitute for it. A qualified clinician's direct measurement remains the authoritative reference.
Surfaces the camera cannot see (the underside of the limb resting on a surface, the deep interior of a cavity) will be missing or interpolated in the mesh.
Known failure modes are documented in the scan's reconstruction report, including insufficient coverage, ambiguous cluster isolation, and low scale confidence.

06 Processing Overview Captured frames upload to a GPU reconstruction pipeline and return a 3D mesh with dimensions.

Capture on the phone produces a set of timestamped frames plus camera trajectory and (optionally) ChArUco detection metadata.
Structure-from-Motion reconstructs the 3D position of every camera and a sparse point cloud.
Dense reconstruction fills in the surface between sparse points to produce a detailed point cloud.
Stump isolation removes the floor, calibration board, and any background clutter so the final mesh contains only the limb.
Metric scaling applies the ChArUco-derived or trajectory-derived scale factor and reports dimensions in mm.
Export to GLB; the clinic dashboard displays the model, dimensions, calibration metadata, and links to intermediate debug artefacts.

Processing takes 20–50 minutes per scan on a mid-range GPU worker and is fully automated — the clinician receives a notification when the model is ready.

07 Data & Privacy Scan data is stored encrypted, access-controlled, and visible only to the clinic it was captured for.

End-to-end TLS for every upload and download; no scan data traverses the network unencrypted.
Storage encryption at rest. All frames, meshes, and reports are held in Firebase Storage with Google's managed encryption keys.
Per-clinic access control. Scans are visible only to authenticated members of the clinic that captured them; patient-level access is gated through an explicit consent flow.
Data minimisation. Raw frames are retained for pipeline reprocessing; they are not shared with third parties and are not used for training without specific, separately-obtained consent.
Patient rights. Patients may request export or deletion of their data through the clinic that captured it.

08 Use Cases & Clinical Context Intended to support — not replace — clinical assessment workflows.

Early documentation of residual-limb shape, captured at the point of first consultation and attached to the patient record.
Referral preparation. A prosthetist receiving a referral can inspect geometry and approximate dimensions before the patient arrives.
Longitudinal tracking of change over time (post-operative, during maturation, after socket fitting) using repeatable smartphone capture rather than recalling the patient for studio scanning.
Research and teaching — anonymised reconstructions support training material and retrospective shape analyses.
Not a replacement for clinical measurement, cast-based socket fabrication, or professional fitting. Augmene is a supplementary digital record.

A short guide for patients, clinicians, and research partners