Evaluating robustness of gait event detection based on machine learning and natural sensors

• Published in: IEEE transactions on neural systems and rehabilitation engineering Vol. 12; no. 1; pp. 81 - 88
• Main Authors: Hansen, M., Haugland, M.K., Sinkjaer, T.
• Format: Journal Article
• Language: English
• Published: United States IEEE 01.03.2004; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Action Potentials; Adult; Algorithms; Artificial Intelligence; Control systems; Cuffs; Electric Stimulation Therapy - methods; Electrodes; Electromyography - methods; Event detection; Female; Gait; Gait Disorders, Neurologic - diagnosis; Gait Disorders, Neurologic - etiology; Gait Disorders, Neurologic - physiopathology; Gait Disorders, Neurologic - rehabilitation; Hemiplegia - complications; Hemiplegia - diagnosis; Hemiplegia - physiopathology; Hemiplegia - rehabilitation; Humans; Leg - innervation; Leg - physiopathology; Machine learning; Muscle, Skeletal - innervation; Muscle, Skeletal - physiopathology; Neuromuscular stimulation; Pattern Recognition, Automated; Peroneal Nerve - physiopathology; Proprioception; Quality Control; Real time systems; Recording; Reproducibility of Results; Robustness; Sensitivity and Specificity; Sensor systems; Sensors; Shoes; Stimulation; Sural Nerve - physiopathology; System testing; Time measurements; Timing; Training; Treatment Outcome; Wear
• ISSN: 1534-4320; 1558-0210
• DOI: 10.1109/TNSRE.2003.819890

A real-time system for deriving timing control for functional electrical stimulation for foot-drop correction, using peripheral nerve activity as a sensor input, was tested for reliability to investigate the potential for clinical use. The system, which was previously reported on, was tested on a hemiplegic subject instrumented with a recording cuff electrode on the Sural nerve, and a stimulation cuff electrode on the Peroneal cuff. Implanted devices enabled recording and stimulation through telelinks. An input domain was derived from the recorded electroneurogram and fed to a detection algorithm based on an adaptive logic network for controlling the stimulation timing. The reliability was tested by letting the subject wear different foot wear and walk on different surfaces than when the training data was recorded. The detection system was also evaluated several months after training. The detection system proved able to successfully detect when walking with different footwear on varying surfaces up to 374 days after training, and thereby showed great potential for being clinically useful.