We used machine learning and survey results from individuals with Parkinson’s disease about their preferences on movement sensor or inertial measurement unit (IMU) placement to develop optimal sensor combinations and algorithms for freezing of gait detection.
- A convolutional neural network detects clinically relevant freezing of gait metrics from raw IMU data
- A single ankle IMU performs nearly as well as three IMUs on the lumbar and ankles
- The lumbar and ankle IMUs were rated highly wearable by patients
We share our data and software openly at: https://github.com/stanfordnmbl/imu-fog-detection