Preliminary development and validation of a textile-based pressure-sensing system for lower-limb prosthetic sockets

Abstract

Background

The socket is an essential component of prostheses and critical to the mobility, quality of life, and independence of individuals with lower-limb loss. A poor-fitting socket can lead to residual limb health issues, gait abnormalities, and prosthesis abandonment. To mitigate these risks, prosthetists routinely evaluate socket fit, but current assessments rely largely on subjective measures, which may not consistently ensure optimal outcomes. The use of pressure sensors at the limb–socket interface could enhance these evaluations, yet challenges with practicality and wearability have limited their clinical adoption. To address these gaps, we developed and evaluated a prosthetic sock with integrated textile-based pressure-sensing cells.

Results

Individual cells were first loaded to evaluate sensitivity, repeatability, and drift. Then, a surrogate residual limb was fabricated from silicone and 3D-printed components to assess performance at the limb–socket interface. The sensitivity test demonstrated the ability to detect load changes equivalent to 5% of the maximum load throughout the full range, while the repeatability evaluation resulted in an intraclass correlation coefficient of 0.998 (0.996–1.000) and a percentage coefficient of variation (CV) below 2%. The average drift over 10 min was 3.03 ± 0.44%. When loading the residual limb in the neutral standing position, three out of four sock cells tested at the limb–socket interface showed percentage CVs below 10%. Finally, all cells tested detected changes in socket pressure distribution under simulated gait events when applying a load of 600 N.

Conclusions

These results were similar to commonly used off-the-shelf pressure sensors, and the sock performance is, therefore, promising. However, further development is required, including form factor improvements, implementation of shear-sensing capabilities, calibration, and validation with prosthesis users.

Background

Amputations can be caused by a range of conditions, including cancer, diabetes, trauma, and vascular disease [1]. When a prosthesis is not regularly used, lower-limb amputations, which are the most common, can have a direct impact on the physical and mental health of the individual. These effects include a decrease in social and physical activities, an increase in pain interfering with daily living and a greater risk for depression [1, 2]. Prostheses are, therefore, essential for the mobility, independence, and quality of life of individuals with lower-limb loss [3]. Within the prosthetic system, the socket is a critical component, facilitating interaction and enabling the transfer of loads and movements between the residual limb and device [4].

Proper prosthetic socket fit is important to the general health and rehabilitation of lower-limb prosthesis users (LLPUs) [5]. A loose-fitting socket can cause pistoning, which is the vertical movement of the socket with respect to the limb, leading to friction and tissue irritation [4, 6]. When the socket is overly constricting, sustained pressure points in the weight-bearing areas of the limb can result in poor blood circulation and the development of pressure ulcers [4, 7, 8]. Poor socket fit can also reduce stability and confidence during gait, increasing the risk of falls [4, 6, 7]. Fit issues and discomfort are major contributors to prosthesis abandonment, with half of LLPUs not wearing their devices on a regular basis [4, 7, 9]. Among those who do wear their devices regularly, more than 50% experience some level of pain [9].

At the socket and limb interface, liners or layers of prosthetic socks are typically worn to adjust the fit, absorb humidity, and improve comfort throughout regular use [10, 11]. However, fit issues are not always resolved and can instead require socket shape alterations. To determine what alterations are needed, prosthetists will typically conduct fit evaluations at various times throughout the life cycle of the prosthesis, including during fabrication, while adapting to a new device, and throughout regular use [12–16].

Fit evaluations rely on verbal feedback from the user about activity levels, pain or pressure points, comfort throughout regular use, and sock layering practices. The prosthetist also examines the residual limb for blanching and tissue damage, and evaluates gait quality [10, 17]. Other techniques include the use of a ball of clay at the distal end of the socket to assess limb support and powder transfer from the socket wall to the limb to map the load-bearing regions [10]. These methods all rely on the interpretation of subjective metrics [17], which can lead to variations in socket alterations between different prosthetists for the same residual limb [18]. In addition, the strong reliance on subjective user feedback can present challenges in accurately and efficiently assessing fit in certain populations, such as young children or those with limited residual limb sensations [10, 17, 19, 20]. These challenges put these populations at greater risk of developing the residual limb health issues associated with poor socket fit [19].

Pressure sensors have been used to study the limb–socket interface with the goal of supplementing these subjective measures with objective data [11, 18, 21, 22]. These sensors measure the pressure exerted on the residual limb by the socket wall, which can be indicative of socket fit [7, 11]. Even though a direct link between the use of pressure sensors and improved patient outcomes has not yet been demonstrated through clinical trials, prosthetists still find the pressure sensor data helpful in making socket alterations [18]. The use of sensor data was also shown to contribute to more consistent socket modifications between different prosthetists and can assist in reducing unwanted pressure in sensitive limb regions [18]. LLPUs have expressed the need for technologies supporting daily prosthetic use (such as auto-adjusting sockets) and systems that can help provide information to a prosthetist about the fit issues that occur outside the clinic [23]. Those advancements can be made possible through the use of pressure sensing at the limb–socket interface [23]. However, pressure sensors have not yet been adopted within standard clinical practice due to several limitations [24].

First, the majority of existing systems have poor spatial resolution and typically do not cover broad areas of the residual limb [12, 25–28], which can limit the assessment of pressure patterns throughout the limb–socket interface [19]. Furthermore, the regions of interest depend on the specific user due to differences in muscle and tissue properties, such as strength and stiffness, and the overall residual limb shape [11, 19]. Others have used pressure-sensing strips to provide fuller coverage of the limb [22, 29]; however, these can still be difficult to use and apply. A more promising solution entails integrating sensor arrays into prosthetic socks or liners [11, 30]. Wheeler et al. have integrated sixteen optical pressure sensors into a liner format fabricated from silicone applied directly on the skin, demonstrating the practicality associated with a cohesive system [30]. However, the spatial resolution is still limited.

Textile-based sensors offer stretchable and breathable materials [31] that are also compatible with integration into these practical form factors [32]. Stretchability is critical to ensure the system conforms to the curvatures of the limb, and breathability helps avoid the accumulation of moisture on the skin, reducing the risks for irritation and infections [33]. Tabor et al. have implemented textile-based sensors into 3× 3 arrays, but the integration of a large grid of these sensors into practical, clinically relevant, and usable form factors, such as socks or liners, has not yet been achieved [32].

To address these gaps, we iteratively developed and evaluated a prosthetic sock with integrated textile-based pressure-sensing cells. The evaluation in this study was completed in two steps:

1. First, the ability of the sensing cells of the sock to detect increments in applied load (sensitivity), while providing precise (repeatability) and stable (drift) measurements, was evaluated. This was assessed by loading an individual cell.

2. Then, the ability of the sensing sock cells to provide precise measurements when used at the limb–socket interface was tested. Each cell should also offer the sensitivity required to detect changes in pressure distribution in the socket under simulated gait events.

Results

For objective 1, individual sock cells were loaded on a silicone surface to evaluate their sensitivity, repeatability, and drift. A sensing pad containing a single textile-based cell and a commercially available SingleTact sensor (PPS UK Limited, Glasgow, UK) were also included for comparison throughout the evaluation. For objective 2, the sensing sock was loaded using a surrogate residual limb fabricated from silicone and 3D-printed components to evaluate the performance of selected cells, along with the SingleTact. Repeatability was evaluated in a neutral standing position, and sensitivity in three simulated gait events (mid-stance, pre-swing, and loading response).

Single-cell test results

Sensitivity

Figure 1a shows the characteristic curves of the sensing pad, selected cells on the sock (Fig. 9), and the SingleTact obtained from the full range (2–40 N) sensitivity test in the single-cell setup. Since the textile-based sensors (pad and sock) are not calibrated, the figure shows the analog-to-digital converter (ADC) output on the left y-axis and the SingleTact force output on the right y-axis. Figure 1b shows the characteristic curves of sock cells 8,4 and 8,5 within the lower range of 0.5–5 N.

Fig. 1.

Sensitivity test results with the single-cell setup a full 2–40 N range results, b reduced 0.5–5 N range results (dashed line indicates range not covered by Fig. 1a)

Fig. 9.

a Anterior view of the surrogate limb with AP and AD cells, b posterior view with PP cell, c socket with AP and AD SingleTacts, d socket with PP SingleTact; (AP: anterior–proximal; AD: anterior–distal; PP: posterior–proximal; sock cells in orange are aligned with a SingleTact sensor)

Repeatability

Table 1 shows the intraclass correlation coefficient (ICC) results from the repeatability evaluation for the sensing pad, one selected cell on the sock (6,11), and the SingleTact. The three sensors satisfy the requirement for the minimum acceptable ICC (ICC > 0.9) [34, 35]. However, the lower bound of the 95% ICC confidence interval for the pad is below 0.9.

Table 1.

ICC test results with the single-cell setup

	Sensing pad	Sensing sock (6,11)	SingleTact
Intraclass Correlation (95% CI)	0.932 (0.808–0.984)	0.998 (0.996–1.000)	0.998 (0.996–1.000)

Figure 2a shows the percentage coefficient of variation (CV) computed at each load across the three trials. When considering all loads, the sensing sock shows a percentage CV significantly lower than the pad (p < 0.05) and equivalent to the SingleTact (p > 0.05).

Fig. 2.

Repeatability test results with the single-cell setup a percentage CV results for all loads across the three trials (* indicates statistical significance due to p < 0.05), b average sensor output for all loads in each trial used to calculate the ICC and CV

Drift

Table 2 shows the drift test results. The sensing sock shows a relative change across the three loads and two trials equivalent to the pad and SingleTact (p > 0.05).

Table 2.

Drift test results with the single-cell setup (* indicates statistical significance due to p < 0.05)

		Sensing pad *		Sensing sock (6,11)		SingleTact *
		Trial 1	Trial 2	Trial 1	Trial 2	Trial 1	Trial 2
% Drift	10 N	2.65%	2.15%	2.96%	2.17%	3.45%	3.70%
	25 N	2.43%	2.57%	3.27%	3.27%	4.94%	3.03%
	40 N	3.10%	2.54%	3.13%	3.37%	4.07%	2.73%
Average % Drift		2.57 ± 0.31%		3.03 ± 0.44%		3.66 ± 0.79%

Surrogate limb test results

Repeatability

Table 3 shows the results of the repeatability evaluation with the surrogate limb in the neutral standing position. Sensor output ranges and percentage CVs across the three trials are shown for the sock cell in the center of the anterior–proximal (AP) region (cell 8,4), the surrounding cells (Fig. 3), and the SingleTact. The CV could not be calculated for the anterior–distal (AD) and posterior–proximal (PP) regions, since the sensor output at 200 N, 400 N, and 600 N is close to zero and consists mostly of noise.

Table 3.

Repeatability test results with the surrogate limb setup

		Inter-trial output min	Inter-trial output max	Inter-trial % CV
Sock Cell 8,4	200 N	11.40	13.19	4.80%
	400 N	15.58	17.54	4.01%
	600 N	16.10	18.44	4.19%
	800 N	17.65	19.42	3.04%
Sock Cell 8,5	200 N	8.02	9.52	5.57%
	400 N	12.10	14.19	4.04%
	600 N	15.02	17.06	3.69%
	800 N	17.63	19.27	2.73%
Sock Cell 9,4	200 N	4.96	8.40	20.06%
	400 N	12.79	16.44	9.40%
	600 N	18.35	21.15	5.52%
	800 N	21.48	23.58	3.33%
Sock Cell 8,3	200 N	19.04	21.40	4.18%
	400 N	27.04	30.10	4.04%
	600 N	30.10	33.33	3.68%
	800 N	31.48	34.92	3.86%
SingleTact	200 N	0.82	1.14	10.33%
	400 N	1.71	2.04	4.44%
	600 N	2.47	2.82	4.64%
	800 N	3.09	3.74	6.07%

ADC output shown for sock cells, force output shown for SingleTact; bolded data indicates overlapping output ranges

Fig. 3.

Cells of interest for the repeatability test with the surrogate limb setup (AP: anterior–proximal)

Sensitivity

Table 4 shows the results of the sensitivity evaluation with the surrogate limb for the three simulated gait events. The results at 600 N are also illustrated in Fig. 4.

Table 4.

Sensitivity test results with the surrogate limb setup

	Mid-stance	Pre-swing	Loading response	Mid-stance	Pre-swing	Loading response	Mid-stance	Pre-swing	Loading response
Anterior–proximal (AP)
Sock cell 8,4			Sock cell 8,5			SingleTact
200 N	14.73	17.28*	0.40*	10.09	11.90*	3.86*	0.81	1.17*	-0.26*
400 N	20.27	22.48*	1.77*	14.92	19.12*	6.82*	1.80	2.76*	-0.53*
600 N	20.53	27.27*	10.35*	17.71	24.55*	10.88*	2.57	4.50*	-0.72*
	Anterior–distal (AD)
	Sock cell 12,4			Sock cell 12,5			SingleTact
200 N	0.04	0.03	0.23	0.03	0.03	8.45*	-0.11	-0.09	0.82*
400 N	0.00	0.00	12.09*	6.83	0.05*	20.73*	0.63	0.09*	2.62*
600 N	5.26	0.04*	24.36*	16.46	0.32*	28.81*	1.66	0.63*	4.53*
	Posterior–proximal (PP)
	Sock cell 8,11						SingleTact
200 N	0.08	0.07	15.49*				-0.38	-0.59*	0.94*
400 N	0.75	0.06*	21.36*				-0.08	-0.69*	2.45*
600 N	6.20	4.88*	30.85*				0.30	-0.74*	4.32*

ADC output shown for sock cells, force output shown for SingleTact; * indicates statistically significant difference in output between the mid-stance and pre-swing or loading response conditions due to p < 0.05

Fig. 4.

Sensitivity test results with the surrogate limb setup at 600 N (AP: anterior–proximal; AD: anterior–distal; PP: posterior–proximal)

In the AP region, the two sock cells and the SingleTact show a significant increase in output between the mid-stance and pre-swing conditions. The decrease between the mid-stance and loading response conditions is also significant.

In the AD region at 200 N, the two cells and the SingleTact show no difference in output between the mid-stance and pre-swing conditions, while the difference is significant for cell 12,5 and the SingleTact at 400 N. At 600 N, all sensors show a significant decrease between mid-stance and pre-swing, and an increase between mid-stance and loading response conditions.

In the PP region at both 400 N and 600 N, the sock cell and the SingleTact show a significant increase in output from mid-stance to loading response and a decrease with the pre-swing condition. The difference from mid-stance to pre-swing for the sock cell is not significant at 200 N.

Discussion

This study evaluated the performance of a prosthetic-sensing sock designed to address gaps in wearability and practicality for clinical socket fit assessments. Extensive preliminary testing was initially conducted to refine the protocol and determine appropriate materials for the sock fabrication. Then, through formal testing, the sensing cells of the sock were found to provide good sensitivity to applied loads throughout the full range of 0–40 N, while the repeatability test showed an ICC of 0.998 (0.996–1.000) and a percentage CV below 2% at all loads. The average drift over 10 min under multiple loads was low at 3.03 ± 0.44%. When loading the surrogate limb in the neutral standing position, three out of four sock cells tested in the AP region showed percentage CVs below 10%. Finally, all five cells tested from the AP, AD, and PP regions detected changes in socket pressure distribution under simulated gait events when applying a 600 N load.

Sensitivity

The sensitivity of eight sock cells was first evaluated with the single-cell setup. Capacitive pressure sensor sensitivity is typically calculated as S = ((C₁–C₀)/C₀)/△P [36]. In this equation, C₁ represents the capacitance reading after applying the pressure, C₀ is the capacitance before applying the pressure (no-load output), and △P is the pressure applied (in kPa) [36–38]. There are two main barriers to using this equation in this study to evaluate sensitivity and compare results with the literature. First, the equation uses capacitance values [36] and would, therefore, need to be adapted to use ADC output values instead, which might prevent a direct comparison of results with past studies. Second, the no-load output (C₀) varies between textile-based cells due to different material properties, such as compression, at the start of a test [39]. Using this equation would result in a sensitivity that is dependent on the varying no-load output, introducing bias in the results. This would prevent a comparison between cells and with the literature.

For this reason, sensitivity in this study was evaluated based on the characteristic curves (Fig. 1a) [39]. The curves obtained from the sock show that all cells provide an increase in output for each corresponding 2 N increment in applied load. This is necessary for the cells to differentiate between similar loads. Two of those cells (8,4 and 8,5) were evaluated in a reduced range and also satisfied this criterion (Fig. 1b).

These results indicate that individual cells have good sensitivity, distinguishing load changes of 2 N (5% of the full range [18]). The output ranges obtained at each load from the surrogate limb repeatability test can help determine whether this sensitivity is reflected at the limb–socket interface (Table 3). In this test, cells 8,3 and 8,4 did not demonstrate the required range separation for loads above 400 N, while cells 8,5 and 9,4 did. The SingleTact, installed in the socket directly over cell 8,4, also showed this range separation. There are a few potential explanations for these differences.

First, since all cells demonstrate good sensitivity when evaluated outside the socket, sensitivity differences between cells in this test could be a result of the surrogate limb and socket structures. Even though the two groups of cells are in proximity to each other on the limb, the resulting pressure in each location can still differ due to the irregular shape of the socket. Second, differences between the SingleTact and the 8,3 and 8,4 cells could be caused by their position in the test setup. The SingleTact is mounted on the socket wall, while the sensing sock is donned over the silicone limb and covered with 5 plies of traditional prosthetic socks. The SingleTact is, therefore, loaded on a rigid backing, while the sock cells are not, potentially leading to pressure differences. To confirm this hypothesis, additional testing should be completed with a tighter socket with no additional space for traditional socks between the sensing sock and the inner socket wall. This configuration would allow the sock cells and SingleTact to be evaluated directly against each other.

Conclusions about sensitivity can also be drawn from the surrogate limb test with mid-stance, pre-swing, and loading response conditions, simulating changes in socket pressure distributions (Table 4). All cells and SingleTacts in the three limb regions (AP, AD, and PP) showed a significant output difference between the mid-stance and pre-swing or loading response conditions at 600 N. These results show that at higher loads, the sock can detect changes in pressure distributions in the socket when simulating moments associated with knee flexion and extension. It is notable that, similar to the previous test, cell 8,4 showed no change in output between 400 and 600 N (i.e., lack of sensitivity) in the neutral (mid-stance) position. However, in the pre-swing condition, the pressure increases in the AP region, where this cell is located, causing it to show a difference in output between each load. This observation supports the assumption made previously, where the lack of output range separation in the repeatability test with cells 8,3 and 8,4 was likely a result of the lower pressure.

The sensitivity found through these three tests is necessary for the system to be used in the clinic to identify excessive pressure points in the socket. However, additional testing should be completed with a poor-fitting socket to determine the extent to which these points can be identified.

Repeatability

The sock cell tested through the single-cell repeatability evaluation satisfied the ICC requirement (ICC > 0.9) (Table 1). When considering all eight loads, the CV from the sock is also equivalent to the SingleTact (p > 0.05), providing a first indication of performance in comparison with existing systems (Fig. 2a).

Hamilton et al. indicated that a CV below 10% is adequate for pressure sensors to be used in the clinic [40], which the sock satisfied. This threshold is not specific to the evaluation of pressure sensors for socket fit evaluations, but rather a more general guideline for body/assistive device interfaces. Hamilton et al. also evaluated the repeatability of two resistive-based pressure sensors commonly used in biomedical applications. Through the application of repeated forces between 2 and 10 N under various conditions, the lowest CV obtained was 3.1 ± 1.8% [40]. In the current study, the sock provided a CV below 2% for a wider range of 5–40 N. The sock cell repeatability is, therefore, better than point sensors evaluated in other studies, but this assessment could be confirmed by repeating this test within a lower force range of 2–10 N.

The percentage CVs obtained from the repeatability test with the surrogate limb can be used to assess whether the results from single-cell testing are reflected at the limb–socket interface (Table 3). Although the output from cells 8,3 and 8,4 lacks adequate separation between ranges, the percentage CV from these two cells is still below the 10% clinical threshold and the 5% range obtained from the three sensors through single-cell testing. Cell 8,5 also satisfies these requirements. Cell 9,4 shows a CV at 200 N above the 10% threshold [40]. This outlier could be due to this cell experiencing lower pressure than the other three at 200 N, leading to a less stable output. This lower stability is shown by the wider output range for this cell. However, additional testing with loads at and below 200 N should be completed to confirm this assessment.

Repeatability was, therefore, adequate through both single-cell testing and at the limb–socket interface. This performance is essential to provide precise measurements in clinical settings.

Drift

The single-cell drift evaluation measured the relative change in sensor output after a 10-min loading period. The sock showed a relative change across the three loads and two trials equivalent to both the SingleTact and the pad (p > 0.05) (Table 2). However, the statistical power for this analysis was 78.9%, which is below the standard threshold of 80%, implying that some differences between groups might not have been detected.

Swanson et al. evaluated the drift of two resistive-based pressure sensors (FlexiForce A201 and Interlink 402) by applying four different pressure levels (10 kPa, 50 kPa, 100 kPa, and 200 kPa) for 5 min each [41]. The FlexiForce drift error varied between 5.1 ± 2.2% and 18.5 ± 5.4%, while the Interlink varied between 15.0 ± 3.8% and 30.5 ± 8.6% [41]. With twice the load application period (10 min), the sock cell in this study showed an average relative change of 3.03 ± 0.44%, demonstrating an improved performance compared to commonly used resistive-based options.

Low drift is essential for the system to provide a stable output during clinical fit evaluations. Additional testing should be completed with the surrogate limb to confirm that this performance can also be achieved when loading the full sock at the limb–socket interface.

Limitations and future work

There were a number of limitations associated with the sock prototype. First, motion near where the cables are connected (Fig. 5b) introduced noise in specific sensor rows or columns, indicating a potential problem with the snap connectors. While the noise was eliminated during data collection by carefully managing the connections, future versions of the system, and especially ones that will be tested on individuals, will need to be more robust.

Fig. 5.

a Sensing pad, b sensing sock (red squares are provided for illustration; the sensing cells are embedded within the sock material and are not visible), c SingleTact sensor

The single-cell evaluation showed good sensitivity for all cells tested, but the output range varied between cells for the same set of applied loads (Fig. 1a). Therefore, the uncalibrated output of different cells cannot be compared directly. This problem can be addressed through calibration [12, 42]. An identical response across all cells would facilitate that process, as a calibration curve could be obtained for one cell and then applied to all others. However, previous studies have demonstrated that pressure sensor accuracy is improved when calibration is performed in the same environment in which the sensors will be used [40]. Therefore, future work should focus on calibrating all sock cells at once in a socket using, for example, a liner balloon system [43].

There are many challenges associated with calibration. First, further evaluations are required to determine the impact of extended use and machine washing on the integrity of the sensing cell material [44, 45]. If the material properties change over time, the cell response could be altered, affecting the accuracy and adherence to the initial calibration completed by the manufacturer. In that case, recalibration would be needed on a regular basis throughout the life cycle of the system, which could impact its viability as a clinical tool. A practical, simple, and fast method would need to be developed to perform these recalibrations during clinic appointments. Further research would also need to be conducted to determine the exact frequency at which to recalibrate the system to balance performance and practicality.

Through preliminary testing, sensor conditioning was found to be important to ensure consistent measurements between data collection sessions. The sensing cells are capacitive-based, and therefore, their capacitance measurement is inversely proportional to the gap between the two electrodes forming the cell [31]. A greater material compression at the start of a session reduces the gap between the electrodes, increasing the baseline capacitance. The sensor then builds upon this capacitance during loading, which can result in a shifted output. Conditioning avoids this issue by providing an identical amount of material compression to the sensing cell at the start of each test session. Conditioning will need to be considered during cell calibration. The calibration curve will be assigned to the post-conditioning cell response, implying that from that point onward, optimal accuracy will only be obtained if conditioning is completed before using the sock.

In this study, the surrogate limb sensitivity test provided a first indication of the sock’s ability to detect clinically relevant changes in socket pressure distribution. A preliminary test was also completed to begin expanding this evaluation to more specific socket changes. This test involved a set of applied loads repeated under two conditions. First, the same traditional sock layers (5 plies) used in this study were donned over the sensing sock on the surrogate limb. Then, an extra single-ply sock was added, simulating a tighter-fitting socket [25]. The results from this preliminary test demonstrated the sock’s ability to detect the tighter fit. While this result is promising, this test should be repeated formally in the future and expanded to clinical socket alterations, such as the addition of pads used to relieve pressure in specific anatomical regions.

An important limitation of the surrogate limb fabricated for this study is the lack of bony prominences, which are often high-pressure areas. Future work should evaluate the sock with a surrogate limb made of 3D-printed bone models and silicone to better simulate these anatomical features [18, 46]. Following additional surrogate limb testing, the sock should be evaluated with LLPUs.

A 3D scan of the sensing sock installed on the residual limb was taken before and after completing the testing protocol presented in this study. A visual assessment of these scans did not indicate any sock movement during testing (i.e., rotation or vertical shift). Therefore, the sensing cells of interest can be assumed to have remained in the same anatomical region of the surrogate limb throughout all tests. The dynamic testing to be completed in the future with LLPUs also has the potential to cause sock movement. Therefore, pre- and post-testing scans will be acquired again in future evaluations, and a more quantified scan comparison method will be used to identify any sock movement.

The textile-based cells of the sensing sock are also not designed to measure shear forces, which would be necessary for comprehensive socket fit assessments [4, 25, 47]. Therefore, further development is required to enable each cell to measure both shear and normal forces simultaneously [48].

Finally, form factor improvements are needed, including shape modifications to more closely resemble a traditional sock and the inclusion of an opening at the distal end for use with pin-lock prostheses. In addition, the current hardware platform should be redesigned to be lightweight and compact to maximize the practicality of the system for regular use. The sampling frequency should also be increased from 10 to 100 Hz for dynamic data collection.

Conclusions

This study focused on the development and evaluation of a pressure-sensing prosthetic sock. The sensing cells of the sock provided comparable performance to existing thin-film pressure sensors, both when loaded individually and at the limb–socket interface. These results are promising and demonstrate the potential to eventually develop a clinically viable system. Future work should focus on addressing connector issues, redesigning the hardware platform, calibrating the cells, implementing shear-sensing capabilities, and evaluating performance with LLPUs.

Methods

Instrumentation

Sensors evaluated

All textile-based samples in this study are capacitive-based and were fabricated by Myant Inc. (Toronto, Canada). Limited information is available on materials and fabrication processes due to intellectual property.

Before fabricating the sock, a single textile-based-sensing cell with dimensions of 10 × 10 mm was knitted into a pad format (Fig. 5a). This cell is made of two overlapping squares of conductive textile material. The sensing pad was included in the evaluation as it provided a baseline for the performance of a single cell without external factors present in the more complex sock interface.

Then, a sock prototype with dimensions similar to a commercially available prosthetic sock (1SP1RGSH, Knit-Rite, Kansas City, USA) was fabricated (Fig. 5b). The sock consists of two textile layers. The outer layer contains 14 horizontal strips of conductive textile material, and the inner layer contains 12 vertical strips. The strips are knitted with a width of 10 mm and separated by 10 mm. The overlap of the vertical and horizontal conductive strips forms individual sensing cells with dimensions of 10 × 10 mm. This grid, consisting of a total of 168 cells, is designed to provide complete limb coverage and enable the system to accurately measure pressures without prior knowledge of where pressure points or potential problem areas may be located on the limb [11, 19, 22]. The integration of all sensing cells into a single cohesive unit also facilitates the installation and removal of the system on the residual limb. Each strip is connected to a single snap connector, and a total of 26 individual cables then feed into a hardware platform.

In all tests performed, the sensing sock cells should provide performance equivalent to commonly used off-the-shelf sensors and custom-made systems from the literature. The capacitive-based SingleTact pressure sensor (Fig. 5c) was included in all tests for a direct comparison to commercially available sensors. Previous studies have validated the SingleTact based on repeatability [49], drift [49], and hysteresis [12] using a single-cell test setup. It was also tested at the limb–socket interface with a surrogate residual limb [18] and validated with LLPUs during walking, sit-to-stand, and stair climbing exercises [12]. The model selected (CS15–45N) is calibrated by the manufacturer for measurements within a 0–45 N range, has a thickness of 0.35 mm and a diameter of 15 mm [12]. The sensor was used with SingleTact’s I2C board connected to an Arduino, interfacing with the data acquisition software on a computer.

A single-cell test setup was first used to evaluate the sensitivity, repeatability, and drift of the sensing pad, individual sensing sock cells, and the SingleTact (Objective 1). Then, a surrogate residual limb setup was developed to load the full sensing sock and evaluate the repeatability and sensitivity of selected cells and the SingleTact (Objective 2). The surrogate limb is a first step toward evaluating the sock with LLPUs and enables a consistent evaluation without the external variables involved with study participants, such as fluctuations in limb volume, temperature, humidity, and load magnitude [7, 18]. All tests were completed using a universal testing machine (UTM) (Z010 ProLine, ZwickRoell, Ulm, Germany).

Single-cell test setup

In the single-cell test setup, the sensing pad, individual sensing sock cells, and the SingleTact were installed on a 2 cm-thick Shore 10A silicone layer (Dragon Skin 10 Fast, Smooth-On, Macungie, USA) in the UTM (Fig. 6). The silicone layer was used to simulate the basic mechanical characteristics of soft tissue [40, 42]. The sensors were loaded using a tip 3D-printed (Original Prusa XL, Prusa Research, Prague, Czech Republic) with polylactic acid (PLA) secured to the 100 N load cell (0.5% accuracy above 1 N) of the UTM [40]. Since the load area will be larger than a single cell when using the sensing sock at the limb–socket interface, the loading tip was fabricated with a square contacting area 5 times greater (22.36 × 22.36 mm) than the area of a cell [40].

Fig. 6.

Single-cell test setup in the UTM

Surrogate limb test setup

Following the evaluation with the single-cell test setup, the sock’s performance was assessed using a surrogate transtibial residual limb fabricated from silicone and 3D-printed components [18, 46].

The surrogate limb was fabricated in the following manner. First, a 3D scanner (EinScan H2, Shining3D, Hangzhou, China) was used to obtain a digital model of an existing transtibial plaster cast, which was then modified using Canfit (Qwadra, Mérignac, France) to obtain the dimensions necessary to fit the sensing sock. The modified model was used as a base to design all surrogate limb components. An inner core structure was created by offsetting the surface of the original model inward by 20 mm to allow for an even silicone thickness throughout the surrogate limb. The model was 3D-printed at 60% infill density with polyethylene terephthalate glycol (PETG) (Fig. 7a). A mold was also designed and 3D-printed at 30% infill density with PLA (Fig. 7b). After printing, the inner core was installed in the mold (Fig. 7c), and 4 lbs. of Shore 10A silicone (Dragon Skin 10 Medium, Smooth-On, Macungie, USA) were mixed and poured [18]. The silicone was left to cure and then removed from the mold (Fig. 7d).

Fig. 7.

Transtibial surrogate residual limb components a inner core structure, b silicone mold, c inner core structure installed in the mold for silicone pouring, d surrogate residual limb after unmolding, e surrogate residual limb with socks, f custom prosthetic socket, g socket adapter, h single-axis joint

The original plaster cast from which the digital model was obtained featured socket trim lines drawn by a prosthetist. A total surface bearing socket with a wall thickness of 7 mm was designed in Meshmixer (Autodesk, San Francisco, USA) based on these trim lines by extruding the original limb model to leave a 5 mm gap between the silicone and the inner surface of the socket. This gap provided enough space for the sensing sock, two single-ply, and one 3-ply traditional socks to be donned on the limb before installing the socket (Fig. 7e). The socket was 3D-printed at 60% infill density with PETG (Fig. 7f). This infill density was chosen to ensure adequate strength based on prior studies with 3D-printed transtibial sockets [50]. After printing, a pyramid adapter (Ottobock, Duderstadt, Germany) with adjustable anterior–posterior alignment was attached to the square base of the socket (Fig. 7g). A pylon and foot were secured to the adapter, and a prosthetist adjusted the tilt angle on the socket and ankle adapters to complete a standard bench alignment commonly used in clinical practice. A single-axis joint was also designed to connect the surrogate limb to the UTM load cell. The joint was 3D-printed at 60% infill density with PETG (Fig. 7h). The upper part of the joint was inserted into the 10 kN load cell of the UTM, and the lower part was attached to the inner core of the surrogate limb. The completed setup is shown in Fig. 8.

Fig. 8.

Surrogate residual limb test setup in the UTM

Three anatomically relevant locations were assessed, including the middle of the patellar tendon (anterior–proximal), distal tibia (anterior–distal), and popliteal depression (posterior–proximal) (Fig. 9a, b) [18, 25, 51]. A sensing cell on the sock was identified within each of these locations. A SingleTact sensor (one at each of the three locations) was attached to the inner surface of the socket directly over the selected cells (Fig. 9c, d) [18, 25]. Measurements were also taken from adjacent cells (Fig. 9a) to determine whether similar results can be obtained throughout the same region. All cells are identified by their corresponding grid coordinates on the sock (row, column).

Testing protocol

Table 5 provides a summary of the protocol completed with the single-cell and surrogate limb test setups. The sensing cells of the pad and sock are not yet calibrated. Therefore, in all tests, the output of the data collection hardware’s ADC was used for the analysis. There are additional complexities involved in calibrating textile-based sensors, due to variations in their compression, shape, and mechanical characteristics at the start of a data collection session [32]. Calibration will, therefore, be addressed in future work.

Table 5.

Single-cell and surrogate limb test sequences summary

		Loads applied	Timing	Number of trials	Evaluation metric
Single-cell test setup	Sensitivity	20 loads (2–40 N range)	15 s at target load 10 s at 0 N	1	Characteristic curve
	Repeatability	5 × 8 loads randomized (5–40 N range)	15 s at target load 10 s at 0 N	3	ICC, percentage CV
	Drift	10 N, 25 N, 40 N	10 min. at target load 3 min. at 0 N	2	Percentage relative change
Surrogate limb test setup	Repeatability	5 × 4 loads randomized (200 N, 400 N, 600 N, 800 N)	10 s at target load 10 s at 5 N	3	Percentage CV
	Sensitivity	5 × 3 loads randomized (200 N, 400 N, 600 N)	10 s at target load 10 s at 5 N	3 (one per condition)	N/A

Single-cell test protocol

Before starting each single-cell test sequence (Fig. 10), the sensors were conditioned using three 40 N loads [40]. This conditioning is expected to cause the material to compress, leading to an output offset. All sensor readings used for the analysis are referenced to the first no-load output after conditioning to account for this offset. Figure 11 shows the sensitivity output signal from sock cell 6,11, where the red dotted line shows the output used as reference. Conditioning is a common practice and recommended for thin-film commercial sensors to achieve optimal performance [40].

Fig. 10.

Single-cell test sequences workflow

Fig. 11.

Sensing sock (cell 6,11) sensitivity output signal (green line represents the ADC output of the sensing sock cell; blue line represents the force output of the UTM load cell; red dotted line indicates the post-conditioning no-load output used as reference; gray dotted lines represent periods averaged for load measurements)

A 0–40 N range was selected for all single-cell testing, since the maximum pressure experienced at the limb–socket interface in adult prosthesis users is around 400 kPa [52]. With a textile-sensing cell area of 100 mm², this pressure corresponds to a force of 40 N [37]. All single-cell loads for the sensitivity and repeatability tests were applied for 15 s and followed by 10 s at 0 N [40]. The first 5 s of each load application period were discarded from the sensor’s output to provide the sensor enough time to reach a stable value. An average of the following 5 s was used for the analysis [40].

Sensitivity

Hopkins et al. evaluated the response of pressure sensors by applying pressure in 40 kPa intervals within a range of 0–400 kPa [52]. Armitage et al. suggested that a pressure sensor that can detect a 5% deviation in pressure could identify fluctuations relevant to clinical assessments of the limb-socket interface  [18]. To evaluate performance based on this threshold, a 20 kPa interval was selected, which corresponds to a force of 2 N [37]. The sensitivity sequence, therefore, consisted of loads between 2 and 40 N applied in 2 N increments. A separate sensitivity test was also conducted on two selected sock cells in a reduced range of 0.5–5 N with 0.5 N increments to assess performance at lower loads.

The average output at each load was used to produce a characteristic curve. This characteristic curve should show no negative or zero differences between adjacent increasing load outputs for the sensor to adequately detect load changes of 2 N (5% of the full range [18]).

Repeatability

Repeatability was evaluated by applying 5 repetitions of 8 selected loads between 5 and 40 N in a randomized order, for a total of 40 load applications [40]. This sequence was repeated in three separate trials on each sensor (Fig. 12). Between each trial, the sensing cell of the pad or sock was manipulated to relieve the compression incurred by the loading, and conditioning was repeated. The three trials tested inter-session repeatability.

Fig. 12.

Single-cell repeatability data collection & analysis structure

The ICC [34, 35, 53] and the percentage CV [40, 49, 54, 55] were used as metrics for repeatability. The ICC (2,1) was interpreted as a global metric encompassing all loads across the three trials (Fig. 12). The number of repetitions for each load within the trials was determined based on statistical power requirements for the ICC (expected ICC of 0.98, minimum acceptable ICC of 0.9, significance level of 0.05, and sample size of 8). The sample size corresponds to the number of selected loads in the 5–40 N range. With these requirements, a minimum statistical power of 80% can be achieved by repeating each load 5 times [56]. The percentage CV, calculated as the standard deviation divided by the mean, was used to provide a more detailed analysis of the repeatability for each load across the three trials and formulate explanations for the differences in ICC. The non-parametric Kruskal–Wallis test, combined with Dunn’s test for multiple comparisons, was then used to determine any statistically significant differences between the CVs of the three sensors.

Drift

Drift was evaluated by applying loads of 10 N, 25 N, and 40 N for 10 min each. Each 10-min period was followed by 3 min at 0 N. The full sequence was repeated twice, with the same sensor area manipulation performed between the repeatability trials. The deviation in pressure output when the same load is maintained over time corresponds to the sensor drift [41, 49, 52, 57]. From the data collected, a 5-s window was extracted at the start and end of each load application period. Drift was evaluated based on the relative change between the averages of these start and end windows [49]. From the relative change obtained for each load in two trials, an ANOVA combined with Tukey’s HSD for multiple comparisons was performed to determine any statistically significant differences between the three sensors.

Surrogate limb test protocol

Before starting the repeatability and sensitivity tests with the surrogate limb (Fig. 13), a 1 cm block was placed under the heel of the prosthetic foot to represent a neutral standing position (Fig. 14d), and 5 repetitions of an 800 N load [18] were applied to ensure proper suspension of the socket on the limb. All surrogate limb loads were applied at a rate of 40 mm/min, maintained for 10 s, and then followed by 10 s at 5 N. Within each 10-s load application period, the first 3 s were eliminated from the output of the sensor, and the following 5 s were averaged and used for the analysis.

Fig. 13.

Surrogate limb test sequences workflow

Fig. 14.

a Expected pressure change from mid-stance to pre-swing [58], b expected pressure change from mid-stance to loading response [58], c 3 cm block under the ball of the foot, d 1 cm block under the heel, e 3 cm block under the heel

Repeatability

The repeatability sequence consisted of three conditioning loads of 800 N, followed by 5 repetitions of 200 N, 400 N, 600 N, and 800 N in a randomized order, applied to the surrogate limb in the neutral position (Fig. 14d). The load applied by the UTM through to the top of the surrogate limb actively applies pressure on the limb, while the suspension of the socket on the limb applies pressure passively. The test sequence was, therefore, repeated in three trials. Between each trial, the socket was removed from the limb, reinstalled, and the suspension loads were reapplied. This process removes both the active and passive pressure to reset the measurements. The percentage CV at each load across the three trials was then evaluated for the sensing sock and SingleTact.

Sensitivity

The sensitivity test was used to assess the capacity of the sock’s sensing cells to identify fluctuations in socket pressure distributions under simulated gait events (loading response, mid-stance, and pre-swing) [58]. The knee flexion and extension moments reach their peak in the loading response and pre-swing events, respectively. In mid-stance, the loading is vertical [58]. These three events were simulated by installing a 1 cm block under the heel of the prosthetic foot (mid-stance; Fig. 14d), a 3 cm block under the ball of the foot (pre-swing; Fig. 14c), and a 3 cm block under the heel (loading response; Fig. 14e) [58]. The expected changes in pressure distribution in the socket from the mid-stance to pre-swing or loading response events are shown in Fig. 14a, b [25, 58].

In each condition, a sequence comprising three conditioning loads of 400 N, followed by a randomized list of 5 repetitions of 200 N, 400 N, and 600 N, was applied to the limb. A two-sample t test was used to assess differences in sensor output between the mid-stance and the pre-swing or loading response events.

Author contributions

TD, CCN, and JA developed the study methodology. TD developed the test environments and collected the data. TD, CCN, and JA collaborated on data analysis. TD produced the figures and drafted the manuscript. TD, CCN, and JA collaborated on editing the manuscript. JA supervised the project. All authors read and approved the final manuscript.

Funding

JA received funding from the Natural Sciences and Engineering Research Council of Canada (NSERC) Alliance Grants (ALLRP) (Missions 570717–2021) and the Ontario Research Fund (Research Excellence Program—RE011-059). TD received funding from the Kimel Family Graduate Student Scholarship in Pediatric Disability Research, the Ontario Graduate Scholarship, and the Barbara and Frank Milligan Graduate Fellowships.

Data availability

Data sets were generated and analyzed during the current study. However, the sensing sock from which the data was acquired consists of proprietary technology developed by Myant Inc. The data is therefore not publicly available.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interest

The authors declare no competing interest.