Multimodel Sensor Based Fog Prediction in Parkinson's Disease: A Deep Learning Approach

Freezing of Gaits (FOG) is a major and it is poorly understood symptom of Parkinson's disease (PD). 80% of advanced levels of PD patients experience the FOG. Freezing is patient's inability to move which is a temporary halt particularly in tight walking surroundings that causes motor disability or even fall. Detection of FOG is crucial without any proper examination. Real time EEG, ECG, EMG, acceleration (ACC) and skin conductance (SC) data are extracted from twelve PD patients. Extracted data are preprocessed. To enhance FOG prediction; here deep learning (DL) and signal processing technique is introduced. Multi stage spatial temporal based Conventional Neural Network- Bidirectional Long Short Term Memory (CNN-BiLSTM) algorithm is used for classification by training and testing data set. Attention mechanism is used in the LSTM blocks to select most relevant part of FOG in the data set. In the FOG prediction experimental result, the average accuracy achieved is 95.8%.