Measurement of neurologic function is critical for diagnosis, prognosis, and clinical trials. Wearable inertial measurement units (IMUs) offer a compact and unobtrusive sensing method to capture spatio-temporal gait data in clinic and non-clinic environments. We used validated IMU-based algorithms to extract spatiotemporal and inertial gait features in 53 patients with Hereditary Spastic Paraplegia (HSP) (43 UHSP, 10 CHSP), drawn from a cohort of 69 enrolled subjects after excluding trials with incomplete or corrupted data. Classification models were trained to distinguish between uncomplicated (UHSP) and complicated (CHSP) diagnoses of HSP using two feature sets: i) a spatiotemporal (ST) feature set associated to stride length, width, duration, and speed, and ii) a spatiotemporal and inertial (STI) feature set associated with swing acceleration and angular velocity. STI features showed statistically significant differences between UHSP and CHSP patient cohorts and improved classification performance relative to classifiers trained with ST features alone: a linear support vector machine trained on STI features achieved 86.8% accuracy, 70.0% CHSP recall, and an AUC of 0.77 under leave-one-subject-out cross-validation, compared to 73.6% accuracy and an AUC of 0.65 for the same model trained on ST features only. These findings highlight the utility of wearable IMUs for detailed gait assessment and underscore the potential of STI features to capture gait impairment associated with HSP diagnostic subtype beyond what spatiotemporal metrics alone provide. Given the small, imbalanced complicated-HSP sample (n=10), these estimates carry wide confidence intervals and should be interpreted as preliminary. IMU-based approaches provide a scalable and portable tool for evaluating neurologic gait disorders that merits validation in larger cohorts.
No comments:
Post a Comment