Trip Detection Algorithms for Healthy and Amputee Individuals.

Amputation disrupts normal gait biomechanics, reducing strength, balance, and toe clearance. These factors contribute to an increase in falls in amputees, leading to both physical injury and psychological burden. Powered and microprocessor prosthetic legs can potentially react to external perturbations, but fast and reliable perturbation detection algorithms are needed. We developed three different algorithms for trip detection, each based on a different approach: thresholds, machine learning, or anomaly detection. Each trip detection method was validated offline with data acquired from 17 healthy subjects, 7 transtibial amputees, and 1 transfemoral amputee. Data was acquired using a modular tripping-eliciting platform based on multiple compliant trip blocks that deliver unanticipated tripping perturbations. All the detection approaches achieved satisfactory performances, with median sensitivities above 80% and specificities above 90%, depending on the sensors and feature combination used. Additionally, the algorithms were able to retain the same level of accuracy on the amputees dataset when trained on the healthy data only, suggesting a high generalization capability. The proposed trip detection algorithms demonstrated implementability under realistic operating conditions, achieving detection times below 150 ms. Among them, the threshold algorithm achieved the fastest detection, with an average time of approximately 40 ms. Overall, all the tested approaches proved to be viable solutions for the real-time detection of trip perturbations, with fast detection and high specificity. Additionally, the results of this work suggest a high generalization capability across healthy and amputees.