Upper limb compensatory strategies across tasks with an ultrasound-controlled prosthesis

Abstract

Upper-limb amputation often leads to compensatory trunk and shoulder movements, increasing the risk of secondary musculoskeletal complications. This issue is exacerbated by the limited wrist functionality in current prosthetic systems, which restrict natural movement patterns during daily activities. Here, we investigate how limitations in the degrees of freedom (DoF) of the prosthetic wrist influence compensatory upper body motion during functionally relevant tasks. Three transradial amputees, experienced in the use of both body-powered and myoelectric prostheses, were fitted with custom sockets and a prosthetic hand controlled via a 32-channel A-mode ultrasound interface. Eight able-bodied participants served as controls. All participants performed three standardized tasks-drinking, lightbulb insertion, and the Clothespin Relocation Test-while trunk and shoulder kinematics and kinetics were recorded using motion capture and surface electromyography. The results demonstrated task-dependent compensation, with the Clothespin Relocation Test eliciting the greatest trunk flexion and bilateral shoulder involvement. Distinct adaptive patterns emerged between dominant and non-dominant sides, with increased reliance on proximal joints as wrist DoFs were restricted. The findings highlight the need for prosthetic designs and rehabilitation strategies that are tailored to specific tasks and user movement patterns. Quantifying compensatory motion provides a foundation for developing user-centered control systems that enhance function and reduce long-term musculoskeletal strain.

Introduction

Upper limb amputation profoundly disrupts motor function and often leads to compensatory movements in the trunk and shoulder, increasing the risk of overuse injuries and musculoskeletal complications^1,2. While myoelectric prostheses have advanced in mechanical and control capabilities, they remain constrained by their limited number of degrees of freedom (DoFs), particularly at the wrist, impairing the intuitive execution of complex daily tasks^3,4. As a result, users frequently adopt unnatural movement patterns, relying on proximal joints to compensate for distal limitations^5,6. These adaptations, though functionally beneficial in the short term, can contribute to fatigue, joint strain, and reduced long-term quality of life^7,8.

A major challenge in upper-limb prosthetics research is understanding how compensatory strategies vary with task demands. Previous studies have compared overall kinematic differences between prosthesis users and able-bodied controls^9–12, but the task-specific nature of compensation has received less attention. Our prior work on healthy individuals with restricted wrist mobility demonstrated that compensation is highly dependent on task context: activities such as drinking elicited minimal shoulder and trunk involvement, while tasks like the Clothespin Relocation Test (CPRT) required substantial proximal engagement (unpublished results by the authors). Similarly, other studies have shown that the Southampton Hand Assessment Procedure (SHAP), due to its seated tabletop setup, results in minimal joint excursion compared to the more physically demanding Box-and-Blocks Test (BBT) and CPRT¹³.

Human–machine interface (HMI) technologies for upper-limb prostheses-particularly those based on surface electromyography (sEMG)-have demonstrated effective simultaneous and proportional control of hand and wrist movements using regression-based or musculoskeletal model methods^14,15. However, achieving fine, coordinated control across multiple DoFs, especially at the wrist, remains technically challenging, potentially limiting smooth, natural movement. To explore alternative control paradigms, this study employs an ultrasound-based HMI that tracks muscle deformation to enable position-level control.

Despite growing clinical awareness of compensatory movement in prosthesis users, there remains a lack of systematic biomechanical studies examining how these strategies vary across tasks and with differing levels of prosthetic wrist control. To address this, the present study investigates upper-body compensation in transradial amputees using a US-controlled prosthetic interface. Participants completed three standardized functional tasks-drinking, lightbulb insertion, and the CPRT-while shoulder and trunk kinematics and muscle activity were recorded using motion capture and surface EMG. This study makes three key contributions: (1) it provides a detailed, task-level analysis of compensatory movement patterns in prosthesis users using a multi-DoF US interface; (2) it introduces a novel comparison framework between prosthetic configurations and able-bodied reference data; and (3) it identifies specific joint and muscle adaptations associated with task complexity and control constraints. These findings offer new insights into user-centred prosthesis design aimed at reducing compensatory burden and improving functional performance.

Results

Compensatory behavior was analyzed using two primary metrics: (1) angular range of motion (RoM) of the shoulders and thoracic spine, and (2) root mean square difference (RMSD) of muscle activation. These parameters were evaluated across three standardized tasks-Drinking, Lightbulb Insertion, and the Clothespin Relocation Test (CPRT)-to examine task-dependent compensatory strategies in prosthesis users.

For each wrist configuration-C (Open/Close + Flexion/Extension), B (Open/Close + Pronation/Supination), F (Open/Close + Flexion/Extension + Pronation/Supination), OC (Open/Close only), and H (participant hand)-RoM and RMSD values were compared to those of able-bodied controls. This resulted in five configuration-specific difference profiles (e.g., C vs. Control), analyzed across tasks to assess how compensation varied with functional demands and wrist DoF availability.

Compensatory behavior

Joint kinematics and muscle activation patterns revealed substantial compensatory movement during functional tasks. Across all tasks, participants exhibited asymmetrical shoulder motion, with greater abduction/adduction (Z-axis) on the prosthesis side, particularly during CPRT () and the Drinking task (). During CPRT, prosthesis-side RoM reached , significantly higher than the contralateral side (, ). Conversely, the contralateral shoulder showed more axial rotation (Y-axis), peaking at during CPRT compared to on the prosthesis side () (Fig. 1). These findings suggest compensatory trunk and shoulder rotation on the contralateral side to support postural stability and reach.

Fig. 1.

Shoulder joint compensation across tasks. Angular range of motion (RoM) across six upper-body degrees of freedom during prosthetic task performance. Each subplot represents joint movement along an anatomical axis: X (flexion/extension), Y (internal/external rotation), and Z (abduction/adduction). RoM values are grouped by wrist configuration (C, F, B, OC, H) and compared against the able-bodied control group. Colored markers indicate individual subject data. Asterisks denote statistically significant differences (*, **, ****). Data were collected during three tasks: (A) CPRT, (B) drinking, and (C) lightbulb, using motion capture.

The detailed comparison of compensatory joint kinematics and muscle activity across prosthetic wrist configurations is summarized in Table 1. Here, arrows indicate relative increases () or decreases () in range of motion (RoM) or muscle activity, double arrows (, ) denote pronounced changes, and () indicates balanced motion. This table highlights how different wrist configurations (OC, C, B, F) elicit distinct compensatory strategies relative to healthy controls.

Table 1.

Summary of the compensatory index of joint kinematics and muscle activity across different prosthetic wrist configurations compared with the healthy control. Arrows indicate relative increases () or decreases () in range of motion (RoM) or muscle activity, double arrows (, ) denote pronounced changes, and () indicates balanced motion. Configuration definitions: OC allows opening/closing of the prosthetic hand only; C allows opening/closing and wrist flexion/extension; B allows opening/closing and wrist supination/pronation; F allows opening/closing, wrist supination/pronation, and flexion/extension.

Configurations	Shoulder RoM ()	Trunk RoM ()	RMSD ( V)	Significant
OC vs Healthy Controls	Across axes	Z-axis rotation	Lower Trapezius RMSD (up to 863.2  V)
B vs Healthy Controls	Z-axis	X-axis	Deltoid (22.8  V)
C vs Healthy Controls	Z-axis; Y-axis	Z-axis; Y-axis	Deltoid (22.8  V), Lower Trapezius (469.2  V)
F vs Healthy Controls	Across axes	Thoracic involvement	Stable (<4  V in postural muscles)	n.s.

Lateralized muscle activation mirrored joint asymmetries. During CPRT, the right deltoid (prosthesis side) exhibited high variability (RMSD: 4.7–22.8 V) (Fig. 3), consistent with its role in sustained abduction. The right lower trapezius also showed elevated activation (RMSD: 282.4–469.2 V), reflecting scapular stabilization during overhead reach. On the contralateral side, the left sternocleidomastoid and upper trapezius showed moderate activation (RMSD: –10.7 to –9.3 V and –18.5 to –13.9 V, respectively), indicating compensatory engagement due to trunk and shoulder rotation.

Fig. 3.

Dominant-side muscle activation across tasks. Root mean square difference (RMSD) of surface EMG signals during three tasks-Drinking, Lightbulb, CPRT-under five wrist configurations (OC, B, C, F, H). Each subplot shows RMSD ± 1 SD for six upper-body muscles: sternocleidomastoid, deltoid, upper trapezius, lower trapezius, teres major, and erector spinae. CPRT yielded the highest RMSD, especially in lower trapezius and teres major. Asterisks indicate statistical differences (*, **, ****).

Thoracic mobility followed a similar task-dependent pattern (Fig. 2). CPRT elicited the greatest trunk RoM across all axes. Z-axis rotation peaked at , significantly higher than during the Drinking (, ) and Lightbulb (, ) tasks. Flexion/extension (X-axis) also reached a maximum during CPRT (), exceeding values for Drinking (, ) and Lightbulb (–, ). Y-axis lateral bending was similarly elevated in CPRT (– vs. –, ). These results confirm that CPRT, the most biomechanically demanding task, elicited the highest angular RoM and muscle activation across the trunk and shoulders.

Fig. 2.

Thoracic mobility across tasks. Angular RoM across three thoracic degrees of freedom during prosthetic task performance. Subplots represent movement along anatomical axes: X (lateral flexion), Y (axial rotation), and Z (flexion/extension). RoM values are grouped by wrist configuration and compared to controls. Colored markers indicate individual subject data. Asterisks denote statistical significance (*, **, ****). Tasks: (A) CPRT, (B) drinking, (C) lightbulb.

Influence of wrist DoF configuration

This section evaluates how varying wrist DoFs compensatory strategies in the shoulder and trunk. Statistical analysis revealed that wrist DoF configuration had a significant impact on upper-body compensatory strategies, particularly in the prosthesis side shoulder and thoracic spine. Repeated-measures ANOVA showed a main effect of configuration on prosthesis side shoulder abduction/adduction (Z-axis; ), prosthesis side shoulder rotation (Y-axis; ), and thoracic rotation (Z-axis; ).

The configuration that allowed Open/Close with wrist flexion/extension (C) consistently elicited the greatest compensatory RoM at the prosthesis side shoulder and thorax. Specifically, prosthesis side shoulder Z-axis RoM under this condition was significantly greater than in the combined flexion/extension + pronation/supination condition (F; ), the pronation/supination condition (B; ), and the Open/Close only condition (OC; ). Similarly, prosthesis side shoulder Y-axis RoM under C exceeded that of B (), indicating that more restricted wrist control (C) led to greater compensatory shoulder motion (Fig. 1). In the thoracic region, Z-axis rotation under the C condition was significantly higher than OC () and B (), while Y-axis (lateral bending) was also elevated compared to F (). X-axis (flexion/extension) differences were less pronounced but still significant () (Fig. 2). These findings suggest that reduced wrist DoFs impose a biomechanical burden on proximal joints, driving increased thoracic involvement to compensate for the loss of distal articulation.

Consistent with the kinematic trends, sEMG analysis also revealed configuration-related differences in muscle activity. The Open/Close only configuration (OC) induced marked increases in right lower trapezius RMSD during CPRT and Lightbulb tasks (e.g., up to 863.2 V) (Fig. 3), indicating enhanced scapular activation in response to limited wrist movement. In contrast, the F configuration, offering the most wrist control, was associated with more stable activation in postural muscles such as the sternocleidomastoid and upper trapezius, particularly on the contralateral side (RMSD V), suggesting reduced compensatory demand.

The contralateral shoulder showed fewer significant changes across configurations. Only in the Y-axis (rotation) did the C condition produce significantly greater RoM compared to B (), indicating that the contralateral limb’s role in compensation was more task- or posture-dependent than configuration-driven.

Overall, these results confirm that lower wrist functionality (C, OC) triggers compensatory movement and elevated muscle activity in proximal segments-especially the prosthesis-side shoulder and trunk-as users must coordinate movements without forearm rotation. In contrast, configurations offering greater wrist mobility (F, B) appear to distribute movement more effectively, reducing the burden on larger joints and stabilizing muscle groups.

Task-specific differences

In this section, we compare joint and muscle compensation across individual tasks to assess task-specific biomechanical demands. Task-level comparisons revealed distinct compensatory strategies in both joint kinematics and muscle activation. While the CPRT consistently imposed the highest overall biomechanical demand-evidenced by elevated joint range of motion (RoM) and root mean square difference (RMSD) values-the Drinking and Lightbulb tasks also elicited substantial compensatory movements under specific wrist configurations. Notably, the Drinking task required increased right shoulder abduction/adduction (Z-axis) during configuration B (pronation/supination only), reaching , compared to in the Lightbulb task and in CPRT (Fig. 1). This was accompanied by increased trunk extension (X-axis: ) (Fig. 2) and elevated right deltoid RMSD values up to (Fig. 3), suggesting substantial compensatory upper-limb engagement when wrist motion was limited to a single DoF. Similarly, the Lightbulb task induced higher contralateral shoulder rotation (Y-axis) under configuration C (flexion/extension only), peaking at , while trunk Z-axis rotation reached (Fig. 1). This was reflected in high variability in lower trapezius muscle activity (RMSD: up to ) (Fig. 3) on the right side, indicating dynamic scapular involvement during overhead reaching. These results highlight the critical impact of restricted wrist control on proximal joint loading and compensatory neuromuscular strategies, even in tasks with lower overall complexity.

During CPRT, prosthesis side shoulder abduction/adduction (Z-axis) reached (), and contralateral shoulder rotation (Y-axis) peaked at () (Fig. 1). Trunk rotation (Z-axis) reached the highest observed value across all tasks at (), significantly exceeding those during Drinking () and Lightbulb () tasks (Fig. 2). These extensive joint excursions were accompanied by high muscle activation. On the prosthesis side, the lower trapezius showed the highest RMSD ( to ) (Fig. 3), along with elevated activity in the upper trapezius ( to ) and teres major ( to ), indicating substantial scapulothoracic involvement. On the contralateral side, muscle activity was more stable, though moderate increases were seen in the sternocleidomastoid ( to ) and upper trapezius ( to ), suggesting compensatory stabilization.

The Drinking task elicited moderate RoM values, with prosthesis side shoulder abduction/adduction at and contralateral shoulder rotation at , both significantly lower than CPRT (). Trunk axial rotation remained elevated (), though lower than CPRT. However, under configuration B, the task elicited notable compensatory behavior due to limited wrist DoF. Corresponding muscle activity reflected this: the prosthesis side deltoid showed sustained recruitment (RMSD: to ) (Fig. 3), while most other prosthesis side muscles-including the upper trapezius, sternocleidomastoid, and erector spinae-remained below the threshold, indicating low variability. On the contralateral side, muscle activity was generally low, with most RMSD values under , except for the sternocleidomastoid ( to ), which indicated some stabilization engagement.

The Lightbulb task generated the lowest overall RoM, though some individual joints still showed notable activity. Prosthesis side shoulder abduction/adduction reached , and contralateral shoulder rotation reached , both significantly lower than in CPRT ( and , respectively). Trunk axial rotation remained moderate (), lower than CPRT () but similar to Drinking. Muscle activation patterns again reflected these reduced kinematic demands. On the right, the deltoid showed active engagement ( to ) (Fig. 3), and the lower trapezius showed high variability (up to ) (Fig. 3), indicative of postural adjustments during overhead movement. Other muscles remained near baseline. The contralateral side showed stable activation, with moderate activity in the sternocleidomastoid ( to ) and upper trapezius ( to ), likely reflecting mild compensatory control. This confirms that task complexity plays a central role in shaping both movement amplitude and muscular demand. CPRT, being the most complex and bilateral task, triggered the largest compensatory patterns, while the Lightbulb and Drinking tasks elicited lower, yet joint- and muscle-specific, responses.

Discussion

This study investigated upper-body kinematics and muscle activation patterns in upper limb prosthesis users across three functional tasks, each performed under five prosthetic wrist configurations. By integrating joint range of motion (RoM) analysis with muscle activation variability (RMSD), we aimed to characterize compensatory strategies and evaluate how prosthetic configuration influences motor behavior. These findings extend prior work highlighting the biomechanical complexity of prosthesis use.

Prosthetic wrist configuration had a clear effect on both joint-level behavior and muscle recruitment. Configurations with reduced DoFs (e.g., OC and H) consistently induced larger RoM in the shoulder and trunk, suggesting proximal compensation in the absence of sufficient wrist control. These results are consistent with earlier work demonstrating that limited wrist articulation leads to overuse of the shoulder complex^16–19. More advanced configurations (e.g., F and C) demonstrated variable results: while they supported more natural movement trajectories, they did not uniformly reduce muscle activation variability. In certain scenarios, high wrist functionality-such as that offered by configurations F and B-facilitated greater voluntary engagement, whereas in others, it supported passive stabilization, depending on the user’s movement strategy and the demands of the task. This duality reflects prior observations that increased prosthetic functionality does not automatically translate to reduced effort^8,20 . These nuanced results underline the importance of assessing not just task success or timing, but the qualitative and physiological cost of movement execution^16,21,22.

The nature of the functional task emerged as a major determinant of compensatory movement strategy. The CPRT consistently elicited the most extensive trunk and shoulder involvement, reinforcing the idea that complex, multistep tasks provoke greater biomechanical demand^18,23,24. These patterns align with previous studies showing that trunk and proximal joint compensation increases with task complexity and limited distal articulation^18,25. In contrast, the Drinking and Lightbulb tasks required more localized and repeatable joint strategies. While simpler in structure, they still demonstrated task-specific compensatory features-particularly in shoulder elevation and minor trunk rotation. These results suggest that compensation is not strictly a result of limb absence, but also emerges in response to the motor requirements and spatial characteristics of the task itself^23,25. This finding is consistent with reports that compensation is not uniformly distributed, but dynamically regulated based on task constraints, goal orientation, and available prosthetic function^{8,17,20,26,27}.

sEMG analysis revealed distinct neuromuscular strategies between the dominant and non-dominant sides. On the prosthetic side, high variability in muscle activation-particularly in the lower trapezius-was observed during tasks requiring dynamic positioning and elevation. This aligns with prior studies that identified increased scapular recruitment as a compensatory response to reduced wrist mobility^17,28. Conversely, muscles such as the deltoid and erector spinae remained stable across tasks and configurations, particularly on the contralateral (non-dominant) side, suggesting a supportive or stabilizing role. This asymmetry reflects a functional division of labor, with the prosthetic limb engaged in manipulation and the contralateral limb providing postural control-an observation also reported in stroke populations and other unilateral impairment groups^26,27. Notably, in configurations with enhanced control (e.g., F and B), there was a reduction in upper trapezius and sternocleidomastoid variability, suggesting a drop in compensatory elevation strategies. These findings support the argument that more responsive or supportive configurations can mitigate unnecessary muscular engagement, leading to more efficient control patterns^21,22.

Furthermore, muscle activation required for prosthetic control can influence compensatory movements, with ultrasound-based control demanding lower activation than sEMG-based systems, thereby minimizing unintended co-contraction and stiffness.

The patterns observed across tasks and configurations have several implications for both prosthetic design and rehabilitation. First, the consistent functional asymmetry between limbs highlights the need for training protocols that address overuse and fatigue in the dominant shoulder and trunk^18,18,23,29. These strategies should also promote adaptive engagement of stabilizers on the non-dominant side, particularly in users who rely on their contralateral limb for postural support. Second, prosthetic wrist joint configurations that minimize unnecessary proximal joint activation while still enabling effective distal control may offer the best balance between biomechanical efficiency and cognitive demand. This underscores the value of developing user-adaptive, task-aware systems-prosthetic interfaces capable of modulating control strategies or providing additional support based on the complexity or requirements of the ongoing activity^17,25,26. The observed division of functional roles-dominant shoulder for manipulation and contralateral shoulder for stabilization-reflects typical limb dominance patterns. The consistent increase in lower trapezius activity across tasks highlights the critical role of scapular stabilization during assisted reaching. Importantly, configurations that can reduce postural muscle overuse while maintaining or enhancing activation in dynamic movers may offer the most effective biomechanical assistance for daily function.

Methods

Study participants

Three right-handed individuals with upper-limb loss (all male; mean age: 43.0  19.5 years; mean weight: 71.7  18.4 kg) were recruited through networks, academic institutions, and online outreach platforms. All participants had prior experience using either passive, body-powered, or myoelectric prostheses and were in good general health at the time of the study. Participant TR2 presented with complex limb loss, including right transradial, left transhumeral, and bilateral transfemoral amputations, and completed the study while seated in a wheelchair. Detailed demographic information, including limb deficiency level, amputation cause, prosthesis type, and residual limb length, is provided in Table 2. Additionally, data from seven able-bodied individuals (six males, one female; mean age: 28.2  4.1 years), who completed the same experimental protocol, were included as a control group (P4–P10). Participants were unaware of the specific aims of the study and provided written informed consent in accordance with ethical standards.

Table 2.

Limb loss participant demographic and clinical characteristics.*TR3 demonstrated some pronation/supination despite limited hand degrees of freedom and did not participate using their own prosthesis.

Variable	TR1*	TR2	TR3*
Age (years)	62	44	23
Gender	Male	Male	Male
Dominant hand	Right	Right	Right
Limb deficiency type	Trans-radial	Trans-radial	Trans-metacarpal
Side affected	Left	Right (bilateral TF)	Left
Cause of amputation	Explosion	Sepsis	Congenital
Residual limb length (cm)	18.3	17.0	34.5
Type of prosthesis	Myoelectric	Body-powered	Functional terminal devices
Years of prosthesis use	42	9	23
Experimental note	Only participated in CPRT	Wheelchair user	Bracelet setup used
Training sessions (n)	6	4	2

Experimental setup

Participants were seated in front of a height-adjustable table, with posture standardized across trials. The upper arm was aligned vertically with the torso, and the forearm was positioned parallel to the floor, maintaining elbow flexion at . Table height was adjusted individually to match each participant’s elbow height, following established ergonomic guidelines, and ensuring repeatable and stable posture during task execution. For prosthesis users, the terminal device (TD) was positioned flat on the table surface in a neutral reference posture. The contralateral arm rested on the opposite-side knee to minimize compensatory motion and prevent marker occlusion. Reference joint angles were maintained at for both the elbow and the knee. At the beginning and end of each trial, participants were instructed to place their hand or TD over a predefined marker on the table to ensure consistent initial and final postures across all tasks (Fig. 4).

Fig. 4.

Experimental setup. Start and end position of the terminal device (TD) during task execution. The black cross indicates the standardized position used to initiate and conclude each task. Colored numbers represent the preparation sequence: (1, blue) timer on, (2, purple) task initiation, (3, green) task completion, and (4, red) timer off. The inset shows a participant seated with the prosthetic system prepared for data collection.

Ultrasound-based human–machine interface

The MoUsE (Fraunhofer IBMT, Sulzbach, DE) portable US system was integrated into the prosthetic control setup³⁰ to provide position-level tracking of wrist movements via 32 single-element piezoelectric transducers embedded in a custom socket. Data were processed using previously validated pipelines and machine learning models^31,32 to enable multi-DoF control of the experimental prosthetic wrist. The US interface was employed solely to standardize wrist control across participants and provide a physiologically relevant, low-effort interface for investigating task-dependent compensatory movements, rather than serving as the central focus of the study.

US frames consisted of 1024 echo lines acquired from 32 transmissions, with 32 receive events per transmission (one for each transducer) (Fig. 5A). Reflections were sampled for 110 s at 50 MSPS with 12-bit resolution at an average of 11 frames per second. Lines were time-gain compensated to account for soft tissue attenuation, band-pass filtered between 0.4–1.6 MHz, downsampled by a factor of 10, and Hilbert enveloped.

Fig. 5.

Ultrasound-based control system. (A) A thermoplastic fabricated socket integrated with a 32-channel ultrasound transducer array for muscle deformation sensing. The ultrasound sensors are embedded in a custom socket and secured around the residual limb to enable position-level control of wrist joint movements. (B,C) Participant performing signal training in a virtual environment. Single-element ultrasound transducers were used to control the virtual hand for object interaction tasks, enabling real-time feedback during pre-prosthetic signal control training.

Processed frames were transformed into features by computing the root mean square over 20-sample windows in each line and then reducing to the first 100 principal component features. Training data were collected while participants performed contractions following screen cues, with one training session conducted for each wrist configuration and cues involving individual or combined movement of the active DoFs. A linear regression model was trained for each DoF using the US data and cues. This approach was selected because position-based US control more replicates natural muscle activation patterns, requiring sustained activation to maintain joint positions, and generally demands less physical effort than sEMG. Consequently, unintended co-activation of adjacent muscles is minimized, ensuring that measured compensatory movements primarily reflect task-driven upper-body adaptations.

Socket design and prosthetic hand

A certified prosthetist fabricated a custom thermoplastic socket using standard clinical protocols to match the participant’s residual limb and facilitate optimal transducer placement. The socket included 32 drilled holes in four concentric rings (10, 10, 8, and 4 from proximal to distal), each fitted with a single-element US transducer mounted using a 3D-printed holder (Fig. 5A). The ultrasound sensor setup, comprising lightweight transducers and secured cabling along the residual limb, was configured to minimize interference with natural posture and movement, and no systematic deviations in kinematics attributable to the setup were observed during experimental sessions. A custom version of the Michelangelo Hand (Ottobock, Duderstadt, Germany) (Fig. 6A) with control over the wrist joint flexion-extension was used. The hand open-closing, wrist flexion-extension, and pronation-supination were controlled with direct position control. In some configurations, the hand opening and closing was controlled by a switching mechanism (Fig. 6A) that the user pressed with the contralateral arm. In these scenarios, pressing the switch closed the hand completely while releasing it opened the hand.

Fig. 6.

Sensor setup and functional tasks. (A) Experimental setup integrating the MoUsE portable ultrasound system housed in a 3D-printed casing, an Arduino microcontroller for on/off switch control, and a custom-fabricated thermoplastic socket embedded with 32 ultrasound transducers and fitted with a Michelangelo prosthetic hand. (B) Placement of surface electromyography (sEMG) sensors using the Delsys Trigno Wireless System. Twelve electrodes were positioned bilaterally on the sternocleidomastoid (SCM), upper trapezius, lower trapezius, deltoid, teres major, and lower erector spinae muscles. Electrode placement followed standardized anatomical guidelines to ensure signal reliability and inter-subject consistency^5,7. (C) Definition of the marker set for participants with unilateral upper-limb loss. Reference markers were placed on the C7 vertebra (C7), manubrium (IJ), and bilateral acromion processes (RightAC, LeftAC) to define the thorax segment. The right and left upper limbs were segmented using clusters of markers placed on the scapulae (RightScapSpine, LeftScapSpine), upper arms (RightHumerLatProx, LeftHumerLatProx), and forearms (RightHumerLatDist, LeftHumerLatDist; RElbowM/L, LElbowM/L). Additional anatomical landmarks included the mid- and distal humerus, and lateral and medial aspects of both elbows. (D) Functional tasks performed during the experiment, including the Clothespin Relocation Test (CPRT), Drinking task, and Lightbulb task.

Prosthesis users operated a hand open/close switch with their contralateral limb. This approach decoupled grasp control from the US interface, allowing the user to focus exclusively on wrist-level degrees of freedom, in line with the study’s aim to evaluate compensatory strategies associated with wrist DoF control. Each participant completed three unimanual upper-limb functional tasks at a self-selected pace using five prosthetic wrist configurations; details of the tasks are provided below.

Prosthetic wrist joint DoF configurations

Table 3 outlines the five prosthetic wrist joint configurations evaluated in this study, each varying in the DoFs available at the wrist. All configurations allowed basic hand opening and closing but differed in wrist mobility.

Table 3.

Gesture configurations and corresponding prosthetic degrees of freedom (DoFs).

DoFs	Configuration	P/S	F/E	R/U deviation	Hand grasping
1-DoF	H	Blocked	Blocked	Blocked	Open
1-DoF	OC	Blocked	Blocked	Blocked	Open
2-DoFs	B	Open	Blocked	Blocked	Open
2-DoFs	C	Blocked	Open	Blocked	Open
3-DoFs	F	Open	Open	Blocked	Open
4-DoFs	Natural hand (healthy)	Open	Open	Open	Open

In all prosthetic configurations (H, OC, B, C, F), hand open/close was decoupled to the contralateral limb, and although all tasks (Drinking, Lightbulb, CPRT) required this action, no systematic deviations in posture or upper-limb kinematics attributable to the control scheme were observed.

Configuration H referred to the participant’s own clinical prosthesis, which allowed only grasping without wrist articulation. Configuration OC provided the same functional limitation-hand opening and closing only-but used the experimental prosthetic system. These two conditions were included separately to compare performance between the familiar clinical device (H) and the research setup (OC). The remaining configurations introduced increasing levels of wrist mobility: B allowed pronation/supination, C allowed flexion/extension, and F combined both wrist DoFs. Each configuration was tested independently to isolate its effect on compensatory upper-limb movements. Additionally, eight able-bodied participants completed the same tasks using their natural limb function without constraints, serving as a control group to represent non-compensatory baseline movement.

Ultrasound control system training program

All prosthetic users participated in a structured US-based control system training program, comprising up to six progressive sessions aimed at improving control accuracy, task performance, and functional integration. The first session involved a clinical examination, including patient history, physical assessment, and introduction of a home-based exercise regimen to enhance posture and muscle conditioning-factors known to support prosthesis adaptation^33–36. Control system orientation and baseline self-report surveys were also conducted. The second session introduced basic control exercises using virtual environment feedback (Fig. 5B,C), accompanied by pre- and post-session assessments of pain, fatigue, and cognitive effort. The third session focused on refining control through task repetition and simple target-matching exercises to improve voluntary motor consistency³⁷. In the fourth session, prosthesis fitting and integration training began, emphasizing US-based control strategies and posture awareness to minimize compensatory movements³⁵. The fifth session introduced functional activity training using daily-living-inspired tasks, while the sixth session evaluated compensatory movement patterns during real-world simulations to promote ergonomic usage.

Ideally, the six-session training program was intended to be completed in close succession, without extended breaks between sessions. However, the actual duration and progression varied depending on each participant’s performance during the initial session. Participants who demonstrated sufficient control and engagement were able to continue with minimal delay and, in some cases, skip certain sessions. Others required additional sessions in between to consolidate foundational skills. Overall, the structure of the training protocol reflects recommended multistage approaches that include physical preparation^1,34, control development^35,37, and functional integration to optimize prosthesis use^35,38.

Data acquisition

Hardware systems

Optical motion capture: upper-body kinematics were recorded using the Vicon Nexus^TM v2.15 motion analysis system (Oxford Metrics Ltd, UK), which incorporated a 28-camera setup (16 Vantage 8 and 12 Vero v2.2 units). A total of 24 passive reflective markers were affixed to anatomically defined landmarks to capture joint and segment motion (Fig. 6C). Marker placement followed standard biomechanical conventions, including key points on the trunk (e.g., sternal notch³⁸, xiphoid process¹¹, C7 and T8 vertebrae), shoulder girdle (e.g., sternoclavicular⁸, acromioclavicular joints¹⁴, acromion angle^23,39), scapula (ScapSpine), and upper limb segments (proximal/distal humerus, medial/lateral epicondyles). For prosthesis users, equivalent marker placements were adapted to the socket and prosthetic components to mirror anatomical references. Motion data were captured at a sampling rate of 200 Hz.

Surface electromyography system: trunk and shoulder muscle activity was captured using the Delsys Trigno Wireless surface EMG system (Delsys, Boston, MA, USA). The system operates with a bandwidth of 20–450 Hz and supports a maximum sampling rate of 2000 Hz, utilizing dual onboard reference electrodes for signal stability. A total of twelve EMG sensors were bilaterally placed on key postural and upper-limb muscles: sternocleidomastoid (SCM), upper trapezius, lower trapezius, deltoid, teres major, and lower erector spinae. Sensor placement followed established anatomical guidelines^1,7,8,13 and was applied symmetrically on both sides of the body to facilitate comparative analysis (Fig. 6B).

Biomechanical model and analysis

A biomechanical model adapted from the University of Southampton’s upper-limb kinematic framework was used to evaluate shoulder joint motion, specifically glenohumeral (GH) dynamics^16,26,40. The model was extended to incorporate sEMG signals, enabling synchronized analysis of both joint kinematics and muscular activity. All adaptations followed the International Society of Biomechanics (ISB) recommendations for upper-body motion analysis¹⁴.

Kinematic data processing: motion capture data were processed using custom Python scripts¹⁷. Raw trajectories were labelled, interpolated to address missing data, and low-pass filtered using a second-order zero-lag Butterworth filter (cutoff frequency: 6 Hz). Joint angles for shoulder DoFs-flexion/extension, abduction/adduction, and axial rotation-were computed using an X-Z-Y Euler angle sequence^5,22,27,41. To define the shoulder joint center (SJC), the Plug-in Gait model’s chord function was applied. The clavicular segment was constructed from the thoracic origin and shoulder wand direction, while the humerus segment was defined from the SJC to the elbow joint center (EJC). using the following equations:

The origin of the clavicle coordinate system (1) was defined as:

1

where is the position of the clavicle marker, is the unit vector along the clavicle X-direction, and MarkerDiameter is the physical diameter of the marker.

The SJC (2) was calculated using the chord function:

2

where is the left shoulder marker position, is the temporary clavicle Y-direction, and LSOffset is the anatomical offset used in the model.

To define the local coordinate system of the clavicle (3), the following cross-product operations were applied:

3

Finally, the humeral axis (4) was defined as:

4

where EJC is the elbow joint center.

Muscle activity analysis: surface EMG (sEMG) signals were band-pass filtered between 10–500 Hz and processed using a root mean square (RMS) envelope¹⁰. To assess variability in muscle activation across the three functional tasks-Drinking, Lightbulb, and CPRT-the root mean square difference (RMSD, in V) was calculated. RMSD values below 4 V were considered negligible and treated as baseline noise. Results are reported separately for the right and left sides and for key upper-body muscles associated with trunk and shoulder stabilization.

Control group reference: both kinematic and sEMG datasets were compared against a control group consisting of ten healthy participants. Control data were obtained from a previous study conducted by our research group, where participants followed the same experimental and analytical protocol.

Functional tasks

To effectively assess upper limb (UL) function, tasks must balance clinical relevance and repeatability. Highly constrained tasks enable precise identification and quantification of deviations from normative performance but may lack clinical applicability. Conversely, clinically relevant tasks often lack sufficient standardization, reducing repeatability across subjects and test sessions. Therefore, UL tasks should be both goal-oriented and standardized to ensure consistent performance⁴². Unconstrained tasks present challenges in evaluating interventions, as execution can vary significantly, and individuals may employ different strategies to achieve the same goal^2,39. A diverse set of constrained or goal-oriented tasks, incorporating specific setup arrangements and protocols, provides a more comprehensive evaluation of UL function^41,43. Following these recommendations, a combination of goal-oriented and standardized tasks was selected. These tasks align with the WHO-ICF model⁴⁴ and incorporate guidelines from the ULPOM group⁴⁵. The selected tasks required engagement of all three wrist-joint movement components: flexion/extension, supination/pronation, and ulnar/radial deviation. The tasks included:

• Drinking task: the participant raised a cup to their mouth, tilted it approximately 90 degrees as if drinking, and then returned the cup to the starting position.

• Lightbulb task: the participant picked up a light bulb from a table and screwed it into a standing lamp.

• CPRT: the participant moved three clothespins in two phases. In Phase 1, clothespins were moved from the inside to the outside at three different heights, starting from the lowest to the highest level. In Phase 2, the clothespins were moved from the outside to the inside, starting from the highest to the lowest level.

These tasks were chosen to provide a comprehensive evaluation of UL function while ensuring clinical relevance and repeatability (Fig. 6D).

Statistical analysis

All quantitative data were analyzed to evaluate the effect of prosthetic wrist degrees of freedom (DoF) configurations on compensatory movement patterns in the upper body. A repeated-measures analysis of variance (ANOVA) was employed to assess within-subject differences in the mean angular range of motion (RoM) across the five wrist configurations (C, F, B, OC, H). Bonferroni-adjusted post hoc pairwise comparisons were conducted to correct for multiple testing.

Between-group comparisons, examining each prosthetic configuration relative to the control group, were performed using non-parametric and resampling-based methods suitable for the independent group structure and small sample size. Specifically, Mann–Whitney U tests and permutation-based analyses were used to evaluate differences between groups. Effect sizes and 95% bootstrap confidence intervals were reported alongside medians and interquartile ranges to emphasize descriptive and practical significance in addition to statistical inference.

The primary outcome measures included the angular RoM across three anatomical planes-sagittal (X), transverse (Y), and frontal (Z)-as well as the root-mean-square (RMS) values of surface electromyography (sEMG) signals recorded from shoulder and trunk muscles. For participants with multiple repetitions of the same task, a single representative trial was selected based on task completeness and signal quality.

All analyses were conducted using MATLAB (The MathWorks Inc.) and SPSS (IBM Corp., Version XX). Levels of significance were reported as follows: significant (), moderately significant (), or highly significant ().

Conclusion

This study revealed distinct changes in kinematic and muscle activation patterns across the Drinking, Lightbulb, and CPRT tasks, demonstrating that increased task complexity amplifies compensatory strategies. The CPRT, as the most demanding task, required greater joint mobility-particularly in right shoulder abduction/adduction and left shoulder rotation-and elicited the highest activation in the Lower Trapezius, emphasizing its role in postural stabilization and dynamic arm coordination. In contrast, the Drinking and Lightbulb tasks involved lower ROM and muscle engagement, with the Deltoid and Lower Trapezius being most active. The Teres Major and Erector Spinae showed reduced activation in the CPRT, indicating task-specific muscle recruitment, while consistently lower activation in the left Sternocleidomastoid aligned with asymmetrical shoulder movement. These findings highlight the close relationship between task complexity, joint mobility, and neuromuscular demands, offering valuable insights for optimizing prosthetic design and rehabilitation strategies to improve functional outcomes and reduce compensatory strain.

Limitations

Some limitations are of note: the small sample of prosthesis users (n = 3) restricts statistical power and generalizability, and inter-individual differences may have influenced compensatory strategies. The selected functional tasks, while diverse, do not capture the full range of daily activities, and the controlled laboratory setup may not fully reflect real-world movement patterns. Additionally, the ultrasound-based wrist control, as a novel interface, may not directly represent commonly used clinical prostheses, potentially affecting user adaptation.

Despite these constraints, the study provides detailed, task-specific insights into upper-limb compensatory strategies across prosthetic wrist configurations, informing future research and prosthesis design.

Author contributions

HHH designed the study, led data collection and analysis, and was the primary author of the manuscript. BGS developed and implemented the ultrasound interface, and supported manuscript editing. MSB contributed to the initial biomechanical modele. DF and AHM provided supervision, guided data interpretation, and offered critical manuscript revisions, secured ethical approval and funding, and contributed to the study design and final manuscript review. All authors read and approved the final manuscript. All participants signed informed consent forms for participation and publication in accordance with ethical standards.

Funding

This work was supported in part by King Abdulaziz University under Grant I36115, and in part by the EU Commission under the project SOMA (H2020-FETOPEN-2019-899822): Ultrasound peripheral interface and in-vitro model of human somatosensory system and muscles for motor decoding and restoration of somatic sensations in amputees.

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

All participants provided written informed consent after receiving full information about the study’s purpose, procedures, potential risks, and their right to withdraw at any time. All methods were carried out in accordance with relevant guidelines and regulations. Ethical approval was obtained from the Imperial College Research Ethics Committee (ICREC; references: 18IC4685, 22IC7602).