LSTM and Convolution Networks exploration for Parkinson’s Diagnosis

Parkinson’s disease (PD) is the fastest growing neurological disorder worldwide. PD has a huge impact on the patients quality life due to motor alterations and behavioral changes. The advancements in low-cost RGB-D cameras, such as MS Kinect®, generates the possibility to use low-cost devices to obtain motion data, and perform common PD test like gait analysis. In this research project, we explore the use of LSTM and one-dimensional convolutional neural network as a complement for clinical PD diagnose, this could be used to help the doctors and specialist in the complex process of objective PD diagnosis. For this, we automatically extracted features and time patterns of these signals, then we performed some deep learning models and as the main result, Conv LSTM model achieved an 83% prediction accuracy, an 83.5% precision, and 83.4% recall, being able to differentiate between PD and Non-PD gait samples.