Prediction and Detection of Freezing of Gait in Parkinson Patients using 3D Accelerometer Data

Freezing of Gait (FOG) is a disabling motor syndrome affecting the mobility of individuals suffering from Parkinson's Disease (PD). It increases the risk of falls, thereby impairing the quality of life of PD patients. Wearable 3D lower-back sensors can record and measure time-series data during most of the daily activities of a patient. Such time series data can be used for prediction of Freezing of Gait. This paper focuses on the application of two deep learning models: Transformer encoder and BiLSTM architecture in identifying and predicting the Freezing of Gait (FOG) episodes in patients affected by PD using the dataset provided by Michael J. Fox Foundation for Parkinsons Research. The data considered here has been recorded both in a laboratory setting and in a patient's home, allowing it to capture various scenarios for the training of the detection model. Results obtained, will prove vital support in terms of early detection and differential treatment strategies, thus ultimately reducing the impact of FOG on patients' lives.