Balance Mat (BM): Portable Sway Assessment

• Balance Mat (BM) is a portable pressure-mapping device that uses an optical fibre grid to measure postural sway as an accessible alternative to laboratory force plates.
• It employs a 32×40 sensor array, digitizing data at 40 Hz to compute centre-of-pressure metrics such as path and range with robust postural analysis.
• Robot and human studies demonstrate that, when calibrated, the BM yields highly reliable and valid sway measurements in both single- and double-leg stances.

Balance Mat (BM) is a novel device for postural sway assessment developed as a low-cost portable alternative to laboratory force plates (FP), which are considered the gold standard postural sway assessment method in laboratory conditions but are limited by lack of portability and the requirement of high-level expertise. The BM uses optical fibre technology to derive centre-of-pressure (CoP) time series and associated sway metrics, and its evaluation was reported through a combined robot study and human study designed to assess reliability, sensitivity, validity, agreement with FP, and the effect of calibration on inter-device alignment (Shrestha et al., 25 Jul 2025).

1. Device definition and sensing architecture

The BM is a fully portable mat measuring 600 × 700 × 6 mm, with a mass of 2.5 kg, and is approved by Australia’s TGA. Underneath the vinyl top-surface lies a 32 × 40 array of plastic optical-fibre cross-over points, with each crossing forming a discrete pressure-sensitive element. When load is applied, the fibres at each crossing compress, modulating the intensity of light transmitted through that fibre. A microcontroller embedded at one corner measures the change in light intensity as a voltage signal, digitises it at 40 Hz, and streams it via USB to a host computer without needing an external power supply (Shrestha et al., 25 Jul 2025).

This architecture defines the BM as a pressure-mapping platform rather than a direct force transducer. A plausible implication is that its portability and self-contained digitisation are central to its intended use outside laboratory FP workflows, but the pressure-derived nature of the signal also makes calibration consequential for agreement with FP-derived quantities.

2. Host-side processing and CoP reconstruction

The BM firmware handles data acquisition and digitisation at 40 Hz, after which host-side processing proceeds through a defined software pipeline. A short dark calibration run with no load is first used to remove baseline offsets from each channel, and a gain factor is applied to convert raw voltages into proportional pressure units in arbitrary units. At each time-step $t$, the 32 × 40 matrix of sensor readings is assembled (Shrestha et al., 25 Jul 2025).

CoP is then computed as a weighted centroid. Let $(x_i,y_i)$ be the known physical coordinates of sensor $i$ on the mat. Then

$$\mathrm{CoP}_x(t) = \frac{\sum_i P_i(t)\cdot x_i}{\sum_i P_i(t)}$$

and

$$\mathrm{CoP}_y(t) = \frac{\sum_i P_i(t)\cdot y_i}{\sum_i P_i(t)}.$$

These two time series, with $\mathrm{CoP}_{AP}\equiv \mathrm{CoP}_y$ and $\mathrm{CoP}_{ML}\equiv \mathrm{CoP}_x$, form the basis for all subsequent sway metrics. An optional low-pass filter, for example a 5 Hz Butterworth filter, can be applied to remove high-frequency noise, after which the CoP vectors are passed to sway-metric routines (Shrestha et al., 25 Jul 2025).

The processing chain is therefore explicitly CoP-centric. This suggests that the BM is intended to interoperate methodologically with standard stabilometric analyses, even though its raw sensing substrate differs from FP instrumentation.

3. Sway metrics and formal definitions

The BM evaluation used a standard set of postural sway metrics computed from the CoP time series. Let $N$ be the total number of samples in a trial, $t_k = k\cdot \Delta t$ with $\Delta t = 1/40\ \mathrm{s}$, and $(x_i,y_i)$0. Denote the mean of $(x_i,y_i)$1 by $(x_i,y_i)$2 and of $(x_i,y_i)$3 by $(x_i,y_i)$4 (Shrestha et al., 25 Jul 2025).

Sway mean, per axis, is defined as

$(x_i,y_i)$5

Sway absolute mean, expressed as mean absolute deviation, is

$(x_i,y_i)$6

Sway root mean square is

$(x_i,y_i)$7

Sway path, or total excursion, is

$(x_i,y_i)$8

Sway range is

$(x_i,y_i)$9

Sway velocity is defined from path and trial duration, with $i$0 (Shrestha et al., 25 Jul 2025).

In the correlation analyses, FP metrics were also expressed as $i$1, $i$2, and $i$3, where resultant distance was given as $i$4. The metric set therefore spans central tendency, dispersion, excursion, extrema, and temporalized excursion, allowing comparison of BM against FP across distinct stabilometric descriptors.

4. Reliability framework and robot-based test–retest evaluation

Reliability was evaluated using the two-way mixed-effects, single-measure, consistency ICC, often denoted $i$5:

$i$6

Here $i$7 and $i$8 are the mean squares from a two-way ANOVA, and $i$9 is the number of repeated trials. Interpretation followed Koo & Li 2016: ICC < 0.50 poor; 0.50 – 0.75 moderate; 0.75 – 0.90 good; and > 0.90 excellent (Shrestha et al., 25 Jul 2025).

The robot study used a UR10 robotic arm to obtain controlled sway patterns and included masses from 10–110 kg. Test–retest ICC values were reported for double-leg and single-leg stances. In double-leg stance, single-measure ICCs were 0.981 for Mean, 0.987 for RMS, 0.981 for Path, 0.991 for Range, and 0.981 for Velocity; average-measure ICCs ranged from 0.991 to 0.995. In single-leg stance, single-measure ICCs were 0.808 for Mean, 0.808 for RMS, 0.808 for Path, 0.845 for Range, and 0.808 for Velocity; average-measure ICCs ranged from 0.894 to 0.916 (Shrestha et al., 25 Jul 2025).

At the study-summary level, the robot results were reported as test–retest ICC $$\mathrm{CoP}_x(t) = \frac{\sum_i P_i(t)\cdot x_i}{\sum_i P_i(t)}$$0 in double-leg and $$\mathrm{CoP}_x(t) = \frac{\sum_i P_i(t)\cdot x_i}{\sum_i P_i(t)}$$1 in single-leg for all metrics, corresponding to good to excellent reliability. Regression of mass versus sway yielded cubic $$\mathrm{CoP}_x(t) = \frac{\sum_i P_i(t)\cdot x_i}{\sum_i P_i(t)}$$2 in both stances, demonstrating sensitivity to load changes. This combination of repeatability and mass sensitivity is central to the BM’s instrumental characterization.

5. Human-study validity against force plate measurements

Validity was assessed in a human study with 51 healthy participants performing balance tasks on the BM in combination with an FP; the detailed summary specifies 7 stance conditions and simultaneous BM+FP measurement. The principal inferential comparison used Spearman’s $$\mathrm{CoP}_x(t) = \frac{\sum_i P_i(t)\cdot x_i}{\sum_i P_i(t)}$$3 between BM-derived and FP-derived sway metrics (Shrestha et al., 25 Jul 2025).

For eyes open, the reported correlations were:

• $$\mathrm{CoP}_x(t) = \frac{\sum_i P_i(t)\cdot x_i}{\sum_i P_i(t)}$$4: BM Mean 0.275, BM RMS 0.287, BM Path 0.730, BM Range 0.641, BM Vel 0.730
• $$\mathrm{CoP}_x(t) = \frac{\sum_i P_i(t)\cdot x_i}{\sum_i P_i(t)}$$5: BM Mean 0.051, BM RMS 0.068, BM Path 0.647, BM Range 0.657, BM Vel 0.647
• $$\mathrm{CoP}_x(t) = \frac{\sum_i P_i(t)\cdot x_i}{\sum_i P_i(t)}$$6: BM Mean 0.322, BM RMS 0.322, BM Path 0.698, BM Range 0.558, BM Vel 0.698

All double-asterisked values were reported with $$\mathrm{CoP}_x(t) = \frac{\sum_i P_i(t)\cdot x_i}{\sum_i P_i(t)}$$7 (Shrestha et al., 25 Jul 2025).

For eyes closed, the correlations were:

• $$\mathrm{CoP}_x(t) = \frac{\sum_i P_i(t)\cdot x_i}{\sum_i P_i(t)}$$8: BM Mean 0.370, BM RMS 0.451, BM Path 0.798, BM Range 0.847, BM Vel 0.798
• $$\mathrm{CoP}_x(t) = \frac{\sum_i P_i(t)\cdot x_i}{\sum_i P_i(t)}$$9: BM Mean 0.439, BM RMS 0.460, BM Path 0.800, BM Range 0.818, BM Vel 0.800
• $$\mathrm{CoP}_y(t) = \frac{\sum_i P_i(t)\cdot y_i}{\sum_i P_i(t)}.$$0: BM Mean 0.405, BM RMS 0.432, BM Path 0.796, BM Range 0.815, BM Vel 0.796

The abstract characterizes the human-study findings as moderate to strong correlations for sway path and range, and the detailed summary notes that eyes-closed correlations rose, with AP range reaching $$\mathrm{CoP}_y(t) = \frac{\sum_i P_i(t)\cdot y_i}{\sum_i P_i(t)}.$$1 (Shrestha et al., 25 Jul 2025).

A concise summary of the two-study validation design is given below.

Study	Experimental basis	Reported outcome
Robot study	UR10 arm with controlled sway; masses 10–110 kg	ICC $$\mathrm{CoP}_y(t) = \frac{\sum_i P_i(t)\cdot y_i}{\sum_i P_i(t)}.$$2 in double-leg and $$\mathrm{CoP}_y(t) = \frac{\sum_i P_i(t)\cdot y_i}{\sum_i P_i(t)}.$$3 in single-leg; cubic $$\mathrm{CoP}_y(t) = \frac{\sum_i P_i(t)\cdot y_i}{\sum_i P_i(t)}.$$4
Human study	51 healthy participants; 7 stance conditions; simultaneous BM+FP	Moderate to strong correlations for sway path and range; higher eyes-closed correlations

These results indicate that validity was metric-dependent. Path and range exhibited the clearest BM–FP correspondence, whereas Mean and RMS showed lower correlations in several eyes-open comparisons.

6. Agreement analysis, proportional bias, and calibration

Agreement between BM and FP was evaluated with Bland–Altman analysis. For each trial $$\mathrm{CoP}_y(t) = \frac{\sum_i P_i(t)\cdot y_i}{\sum_i P_i(t)}.$$5, let $$\mathrm{CoP}_y(t) = \frac{\sum_i P_i(t)\cdot y_i}{\sum_i P_i(t)}.$$6 be the BM-derived metric and $$\mathrm{CoP}_y(t) = \frac{\sum_i P_i(t)\cdot y_i}{\sum_i P_i(t)}.$$7 the FP metric. The difference is $$\mathrm{CoP}_y(t) = \frac{\sum_i P_i(t)\cdot y_i}{\sum_i P_i(t)}.$$8 and the mean is $$\mathrm{CoP}_y(t) = \frac{\sum_i P_i(t)\cdot y_i}{\sum_i P_i(t)}.$$9. Mean bias is $\mathrm{CoP}_{AP}\equiv \mathrm{CoP}_y$0, and limits of agreement are $\mathrm{CoP}_{AP}\equiv \mathrm{CoP}_y$1. Proportional bias is detected by linear regression of $\mathrm{CoP}_{AP}\equiv \mathrm{CoP}_y$2 on $\mathrm{CoP}_{AP}\equiv \mathrm{CoP}_y$3 through

$\mathrm{CoP}_{AP}\equiv \mathrm{CoP}_y$4

with a slope $\mathrm{CoP}_{AP}\equiv \mathrm{CoP}_y$5 significantly different from zero indicating proportional error (Shrestha et al., 25 Jul 2025).

An explicit example was reported for AP-direction sway path in the human study under eyes-open conditions: mean bias $\mathrm{CoP}_{AP}\equiv \mathrm{CoP}_y$6, 95% limits of agreement $\mathrm{CoP}_{AP}\equiv \mathrm{CoP}_y$7, and regression of $\mathrm{CoP}_{AP}\equiv \mathrm{CoP}_y$8 on $\mathrm{CoP}_{AP}\equiv \mathrm{CoP}_y$9 showed a positive slope, indicating that the BM overestimates more at larger sway magnitudes. More generally, pre-calibration Bland–Altman analysis showed consistent BM overestimation with proportional bias (Shrestha et al., 25 Jul 2025).

To correct this, a simple linear calibration was derived by regressing BM metrics onto FP metrics:

$\mathrm{CoP}_{ML}\equiv \mathrm{CoP}_x$0

In practice, the calibrated BM was obtained as

$\mathrm{CoP}_{ML}\equiv \mathrm{CoP}_x$1

The reported human-study coefficients were:

Metric	Calibration relation
AP sway path	$\mathrm{CoP}_{ML}\equiv \mathrm{CoP}_x$2
AP sway range	$\mathrm{CoP}_{ML}\equiv \mathrm{CoP}_x$3
ML sway path	$\mathrm{CoP}_{ML}\equiv \mathrm{CoP}_x$4
ML sway range	$\mathrm{CoP}_{ML}\equiv \mathrm{CoP}_x$5

After applying calibration, Bland–Altman plots showed mean bias approximately 0, narrower limits of agreement, and no significant proportional trend. The detailed results further state that limits of agreement tightened by approximately 50% and that the slope in $\mathrm{CoP}_{ML}\equiv \mathrm{CoP}_x$6 versus $\mathrm{CoP}_{ML}\equiv \mathrm{CoP}_x$7 was no longer present, confirming alignment with FP (Shrestha et al., 25 Jul 2025).

This establishes an important qualification: the BM demonstrated consistent sway measurement across varied stance conditions and established both reliability and validity following appropriate calibration, but pre-calibration outputs exhibited systematic overestimation relative to FP. A plausible implication is that the BM is best interpreted as a calibrated measurement system rather than an uncalibrated drop-in replacement for FP. Within that condition, the reported conclusion is that it provides highly reliable sway measurements with valid agreement to a gold-standard FP in both single- and double-leg conditions, and that its portability, ease of use, and low cost make it a candidate for clinical or home-based balance assessments once the per-metric calibrations are applied (Shrestha et al., 25 Jul 2025).