A data-driven approach to kinematic analytics of spinal motion

Abstract

A common reason for low back pain can be attributed to postural stress. While seated or walking, bad posture can put strain on the spine. Increased stress on the spine may induce tightness and spasms in the lower back muscles and may lead to low back pain. Recognition of unstructured daily activities becomes a more difficult and essential task, as monitoring of daily activities becomes more important, especially for helping sick and elderly people. In this study, we employ deep learning and machine learning methods to study spine motion and postural stress using two sensors attached to lower back of a healthy subject while the subject is performing regular daily activities. A comparison of the accuracy of deep learning and supervised machine learning approaches (Decision Tree, Random Forest, Gradient Boosting, AdaBoost, KNN, Naive Bayes) in identification and labeling of daily activities is provided. In addition, the effective values for hyper parameters of LSTM neural networks have been determined. LSTM neural networks achieved highest accuracy. © 2019 IEEE.