A pipeline for gait analysis through inertial sensor data for neuroprosthetic and rehabilitative settings

Traumatic injuries or pathologies of the nervous system can impair walking, substantially worsening the quality of life in affected patients. As improved rehabilitation techniques and neurotechnologies help restore walking in impaired patients, more sophisticated measurement systems and algorithms need to be developed to perform quantitative assessments. Such recorded data is useful not only during patient rehabilitation but can assist in the timely diagnosis of neurodegenerative diseases, even redefining existing diagnostic pipelines. In this thesis, we conducted gait analysis in healthy subjects (HC), patients who suffered spinal cord injuries (SCI), and patients affected by Parkinson's disease (PD) using inertial sensors, including gyroscopes, accelerometers, and magnetometers (normally referred to as inertial measurement units – IMUs). These sensors are particularly advantageous due to their economical affordability and suitability for use in unstructured clinical and rehabilitative settings. First, we focused on subdividing the gait cycle into steps and their respective stance and swing phases. While this process is often performed manually and thus is extremely time-consuming, here, we proposed a semi-automatic segmentation of gait events based on the structural features of IMU signals. As this semi-automatic routine depends upon several subject-specific parameters, which can vary considerably among patient data, we developed a machine learning framework to identify gait events in both HC and PD subjects. We then computed temporal parameters and demonstrated how they could be used to distinguish between HC and PD subjects. In recent times, the development of omnidirectional treadmills, body weight support systems, and virtual reality environments has significantly transformed traditional physiotherapy. In this study, we demonstrate, using our gait analysis pipeline, that these devices do not substantially alter gait parameters in either HC or PD subjects. While our results are consistent with expectations across various settings, the segmentation of the gait cycle should be validated against more robust data sources. Sensorized insoles available on the market could help in this respect, but they tend to be extremely expensive and cannot be customized on a patient-to-patient basis, taking into account the orthotic assistance that patients are equipped with. Here, we present the development of a low-cost and potentially customizable insole enabling the identification of gait events. Concurrent acquisition of IMU and insole data has allowed for the comparison of gait events detected by both systems. In the future, the insole can be utilized in combination with IMU sensors to enhance the accuracy and comprehensiveness of gait analysis. In conclusion, the thesis demonstrates the potential of IMU sensors in clinical evaluation and monitoring, offers new perspectives for improving diagnostic techniques in the medical field and evaluating new rehabilitative tools, and proposes the use of a sensorized insole for gait event identification.