Topic

Fitness tracking is great for calories and steps, but similar sensors are capable of reporting much more about how we move throughout the day. This is especially important in assessing the quality of movement for those with limited mobility. Doctors often want to know more detail about patient behavior after therapy to select and adjust the appropriate intervention. Using machine learning on wearable accelerometer signals, we estimate the activities patients with incomplete spinal cord injury are performing. By combining windowed classifier estimates over time using a hidden markov model, we show how error rates can be significantly decreased, which brings more detailed assessments of patient activity closer to a clinical reality.