Time Series Data Analysis and Modeling of Machine Learning Methods in Limb Function Assessment

Abstract

In response to the current needs of patients with limb dysfunction, with the goal of safety, real-time, non-invasive, and intelligent rehabilitation assessment, and with limb dysfunction patients as the research object, the author uses intelligent perception technology to obtain rehabilitation data of patients, fully utilizing the advantages of the data itself, and is committed to achieving rehabilitation training and muscle fatigue assessment for limb dysfunction patients. The author developed an assessment model for limb function evaluation using the Dynamic Time Warping K-Nearest Neighbor (DTW-KNN) algorithm and a Long Short-Term Memory (LSTM) neural network-based evaluation model. Based on the experimental findings, it was demonstrated that DTW-KNN effectively categorizes and assesses the rehabilitation motions of upper limbs during elbow flexion under various completion scenarios. Patients have the flexibility to conduct independent and effective upper limb rehabilitation training at home using the upper limb functional rehabilitation system, without any constraints of time or space. By enabling physicians to promptly modify the rehabilitation plan, the system significantly addresses the limitations of conventional upper limb rehabilitation approaches, lowers the medical expenses associated with stroke upper limb rehabilitation, and helps mitigate the shortage of rehabilitation specialists. Utilizing the developed upper limb functional rehabilitation system, the author gathered a set of inertial sensing data on upper limb rehabilitation movements, showcasing prominent temporal features. Consequently, to address this issue, the author proposes the use of Long Short-Term Memory (LSTM) neural network - a recurrent neural network (RNN) with superior temporal data processing capabilities. Based on the multi-dimensional inertial sensing data collected by the upper limb rehabilitation system, the author constructs a recurrent neural network classification model. The model can accurately classify and evaluate different types of upper limb rehabilitation movements under varying completion scenarios. The experimental results indicate that: the overall classification accuracy of DTW-KNN for elbow flexion, elbow flexion and forearm abduction, and shoulder flexion in upper limb rehabilitation movements is 71.8%, 47.9%, and 68.8%, respectively. It was observed that the classification accuracies of LSTM neural network model were 98.2%, 93.3%, and 95.1%, respectively. This marks a notable improvement in the classification accuracy of LSTM neural network model compared to DTW-KNN, with an increase of 26.4%, 45.4%, and 26.3%, respectively. LSTM has a significant advantage over DTW-KNN in terms of classification time, with less classification time.