Clinical use

Emerging technologies

Three trends are reshaping balance assessment: wearable inertial sensors that take posturography into the community, virtual-reality systems that grade visual challenge under software control, and machine-learning models that find pattern in the multivariate sway record.

Trainee

Wearable IMU systems already correlate strongly with force-plate sway in Parkinson's; trunk-mounted accelerometry reveals postural instability before clinical falls and has been validated for longitudinal tracking outside the lab.1,2Smartphone-only systems are a step further out, with regulatory and validation work in progress.

VR-augmented posturography pairs immersive visual flow with a force plate or IMU. It enables graded sensory desensitisation in PPPD and a high-fidelity vestibular rehabilitation environment that is reproducible across sessions.3 Patient engagement and adherence are notably higher than with conventional rehabilitation.

The three lanes

  • Wearable IMU posturography

    What it is
    Accelerometer + gyroscope nodes on trunk and lower limbs (or a smartphone) sample sway at high frequency in any environment.
    Clinical uses
    Community fall-risk screening, longitudinal Parkinson tracking, tele-rehabilitation, ecologically valid balance measurement.
    Evidence
    Strong correlation with force-plate metrics; validated for ISway-style trunk sway and gait variability in Parkinson's.
  • VR / AR-augmented posturography

    What it is
    Headset-delivered optic flow and virtual environments paired with a force plate or IMU; graded sensory perturbation in software.
    Clinical uses
    Visual-dependency assessment in PPPD; gamified, graded vestibular rehabilitation; paediatric and cognitively impaired engagement.
    Evidence
    Promising trials show improved postural metrics and dizziness-handicap scores when added to standard VRT; clinical evidence base growing.
  • AI / machine learning on sway data

    What it is
    Supervised classifiers, dimensionality reduction and predictive models applied to multivariate CDP and IMU recordings.
    Clinical uses
    Phenotyping balance disorders, early-stage disease detection, individualised rehab planning, real-time fall-risk alerts.
    Evidence
    Demonstrated discrimination of Parkinson, MS and vestibular phenotypes; predictive fall models in older adults. Regulatory and validation work ongoing.

Wearable IMU — what it actually streams

A belt-clip IMU samples three axes of body acceleration at ~100 Hz. Switch scenarios below to see how a quiet stance, a steady gait and a sudden trip produce characteristically different X / Y / Z signatures. The patient figure animates from the same signal, so motion and trace stay in sync.

Xmediolateral
Yanteroposterior
Zvertical

Regular gait. Rhythmic mediolateral and vertical signal at ~1 Hz step frequency.

VR-augmented posturography — optic flow drives sway

The slider increases the optic-flow speed in the headset view. In a vision-dependent patient, sway amplitude scales with that flow — exactly the mechanism graded VR therapy uses to desensitise the visual channel in PPPD. Watch the CoP trace fill the envelope as you push the slider toward 100%.

Patient

Headset view

CoP trace (last 6 s)

50%RMS sway: 0.0 units · within physiological range

A vision-dependent patient (or any patient with high visual weighting) will sway more as the visual surround becomes more dynamic. VR therapy uses graded flow to desensitise the visual channel — the same principle, applied therapeutically.

ML classifier — the decision boundary, made visible

Three pre-defined patient clusters (healthy, vestibular, Parkinson) sit in a 2D feature space — sway area on the x-axis, MCT latency on the y. Drag the cross-hair to test any point; a nearest-centroid classifier predicts the class and reports confidence based on the margin to the next-closest centroid. Borderline points are the ones that matter clinically.

HealthyVestibularParkinsonSway area →MCT latency →

Predicted class

Vestibular

Confidence

0%

Ambiguous — almost equidistant to two or more clusters.

Drag the cross-hair anywhere in the feature space, or tap to teleport.

A nearest-centroid sketch of how a deployed classifier might segment the sway-area / MCT-latency plane. Real systems use more features (strategy score, LoS metrics, IMU spectra) and richer models — but the idea is the same: learn the boundary, then place new patients within it.

Where this leaves CDP

CDP is not going away — its controlled-condition standardisation and the depth of its evidence base are unmatched. What is changing is its boundary. The new tools extend balance assessment outwards into time (longitudinal IMU tracking), space (real-world setting), and complexity (multivariate ML on combined CDP + IMU + VR data). Posturography's next decade will be a hybrid: deep booth-based phenotyping anchored to broad in-the-wild monitoring.

What still needs answering

  • Regulatory clearance for AI-driven fall-risk and diagnostic flags.
  • Reimbursement parity between booth CDP and IMU-based assessments.
  • Reproducibility of VR perturbation across headsets and lighting conditions.
  • Patient acceptability of long-term wearable use outside the clinic.