Classification of Gait Phases Using a Shank-Mounted Single IMU Sensor

Human gait parameters reflect physical and psychological health. Gait impairments hinder daily activities and mobility, particularly in those with neurological or musculoskeletal disorders. While methods for estimating temporal parameters exist, there remains a need for a portable, user-friendly, and cost-effective gait analysis device. Therefore, in this study, a feature engineering-based portable gait characterization module is proposed, and a shank-mounted inertial measurement unit (IMU) is utilized for gait phases (GPs) and event detection. The efficiency of the developed module was assessed using data collected from ten healthy subjects walking on a flat surface. A force-sensing resistor (FSR)-equipped instrumented insole was used as a reference system to validate the results obtained from the developed module. The performance is estimated using three different classifiers: support vector machine (SVM), K-nearest neighbor (KNN), and linear discriminant analysis (LDA). For gait event (GE) identifications, the average classification accuracies depicted by SVM, LDA, and KNN classifiers are 95.69% $\pm \; 5.23$ %, 96.64% $\pm \; 5.02$ %, and 93.63% $\pm \; 4.84$ % ( ${p}\text {-value}\lt {0.05}$ ), respectively. In summary, the GEs and gait temporal parameters are reliably estimated using a single IMU with SVM or LDA classifier ( ${p}\text {-value}\gt {0.05}$ ). Additionally, the efficacy of the proposed model for sensor location and subject variability has been evaluated. The performance of LDA and SVM classifier for GP prediction has been found to be invariant ( ${p}\text {-value}\gt {0.05}$ ) toward sensor location and subject variability. Future work will explore diverse walking conditions and test the module on individuals with gait impairments to evaluate its robustness and clinical applicability.