This study evaluated an accessible gait-analysis pipeline that converts orientation outputs from a consumer-grade IMU system (cIMU) into lower-limb joint kinematics and compared those results with kinematics produced by a research-grade IMU system (rIMU). The cIMU workflow used custom Python scripts implementing an established biomechanical convention (Grood–Suntay joint coordinate system) to compute hip, knee, and ankle angles. The rIMU outcomes were derived using the vendor’s proprietary processing software. Both systems were recorded simultaneously with sensors mounted in standardized adjacent positions.
Thirty-two individuals with unilateral transtibial amputation completed five overground level-walking trials while wearing both IMU systems. Sensors for each system were positioned adjacently on the same body segments to reduce confounding from different sensor locations. The study acknowledged unique methodological challenges in prosthesis users, including altered segment coordination, socket-residual limb motion, and limited distal anatomical landmarks that can complicate sensor-to-anatomical reference alignment.
For the cIMU workflow, segment-orientation outputs served as input to custom Python scripts that applied biomechanical transformations to compute bilateral hip, knee, and ankle kinematics using the Grood–Suntay joint coordinate approach. The rIMU system generated kinematic outputs using its proprietary algorithms. The independent cIMU processing aimed to provide a transparent, reproducible option where integrated biomechanical modeling is not available in consumer devices.
Agreement and reliability between workflows and across repeated trials were evaluated using multiple complementary methods: Statistical Parametric Mapping (SPM) to assess gait-cycle waveform differences, Bland–Altman analysis to quantify bias and limits of agreement, intraclass correlation coefficients (ICCs) for test–retest consistency, intra-subject variability measures, and paired comparisons of discrete kinematic features. Analyses were performed across three anatomical planes—sagittal, frontal, and transverse—and across proximal (hip) and distal (knee, ankle) joints.
Repeated-trial consistency was high for hip kinematics across all three anatomical planes and for sagittal-plane knee and ankle kinematics. Average-measure ICCs for these measures ranged from 0.945 to 0.996, indicating strong within-subject reliability across the five trials.
Inter-workflow agreement was strongest in the sagittal-plane. Bland–Altman mean differences for sagittal-plane knee and ankle angles were within 2.5°, indicating low systematic bias between the cIMU and rIMU workflows for primary flexion–extension measures. However, several frontal- and transverse-plane measures—and some distal-joint variables—showed wider limits of agreement, reflecting greater variability and reduced interchangeability between workflows for non-sagittal kinematics.
SPM identified statistically significant waveform differences across multiple gait-cycle phases, particularly in the frontal and transverse planes. Paired comparisons indicated that differences in discrete kinematic measures were concentrated in non-sagittal planes and at distal joints (knee/ankle), rather than in sagittal-plane hip or proximal measures.
The pattern of results suggests that the cIMU workflow reliably captures broad sagittal-plane gait patterns and can provide consistent repeated-trial measures for clinical monitoring of sagittal kinematics in transtibial prosthesis users. Conversely, non-sagittal-plane kinematics and some distal-joint measures are more susceptible to inter-workflow differences, which may reflect sensitivity to small misalignments, prosthesis-mounted sensor placement, socket dynamics, or limitations of sensor fusion and biomechanical modeling in these planes.
Clinicians and researchers should therefore interpret frontal- and transverse-plane outputs from consumer-device workflows with caution, and avoid treating non-sagittal or distal-joint measures from cIMU processing as directly interchangeable with outputs from research-grade systems without careful validation for the intended application.
Where research-grade gait-analysis systems are unavailable, the independently implemented cIMU workflow presented in this study offers a practical option for repeated-trial assessment of core sagittal-plane gait metrics in transtibial amputees. Its strengths include portability, lower cost relative to laboratory systems, and a transparent processing pipeline that can be inspected and reproduced. The workflow is best applied to monitor broad changes in sagittal-plane hip, knee, and ankle function over repeated sessions. For clinical decisions or research questions that depend on frontal- or transverse-plane measures—or on precise distal-joint kinematics—results from the cIMU workflow should be regarded as provisional unless supplementary validation is performed.
The study’s processed outcome dataset and analysis code are publicly available: the processed dataset is archived on figshare, and the analysis code and related materials are available on GitHub and archived in Zenodo. These resources support reproducibility and enable clinicians or investigators to inspect, adapt, or extend the cIMU processing pipeline.
The study used a research-grade IMU workflow as the comparator rather than optical motion capture; the authors note that research-grade IMUs have previously shown acceptable agreement with optical systems in transtibial amputees but are not an independent gold standard. The findings are specific to the sensors, mounting strategy, processing algorithms, and participant sample used in this investigation. Readers seeking additional methodological detail or the full results tables and figures should consult the publicly archived dataset and code made available by the authors.