Research On Wearable Monitoring Method Of Freezing Of Gait

Freezing of gait(FOG)is an occasional lower extremity dyskinesia symptom.Accurate and timely parametric monitoring plays an important role in the clinical diagnosis,condition assessment,and formulation of rehabilitation treatment plans for Parkinson’s disease and other related diseases.Due to the sporadic of FOG and the inability to capture it in the laboratory environment,the current subjective clinical assessment methods relying on scales cannot accurately grasp the severity and timing of FOG,which affects the diagnosis and treatment of diseases,and may even lead to missed or misdiagnosed.Therefore,it is necessary to develop an objective FOG monitoring method that can be integrated into daily life.Both the Inertial Measurement Unit(IMU)and the force-sensitive insoles are ideal devices for monitoring FOG in daily life,but both have poor robustness when used alone.In order to achieve robust and reliable FOG monitoring,this paper adopts the technical route of fusion of IMU signals and plantar pressure distribution signals of force-sensitive insoles.However,in the current study,the IMU-based monitoring method has a cumbersome equipment wearing process,and the insoles-based monitoring method has large data scale and high power consumption,which makes it impossible to achieve long-term monitoring.Therefore,the optimization of the IMU wearing scheme and the data reduction of forcesensitive insoles are prerequisites for multi-modal fusion methods to be integrated into daily life.In addition,the extraction of robust FOG features from the IMU signals and the plantar pressure distribution signals is an important basis for ensuring the robustness and reliability of the fusion method.To this end,we explored a highly robust FOG monitoring method based on the IMU signals and the plantar pressure distribution signals,respectively,before multi-modal fusion.Firstly,the method to improve the robustness of the FOG monitoring model based on the IMU signals was studied,and the IMU wearing scheme was optimized to realize the lightness and ease of use of the IMU device;Secondly,the method to improve the robustness of the FOG monitoring model based on the plantar pressure distribution signals was studied,and the data of the pressure insole was simplified to realize the lightness and ease of use of the pressure insole;Thirdly,since the gait signals obtained by the IMU and the force-sensitive insoles have the characteristics of complementary information,an adaptive weighting method was used to fuse the IMU signals and the plantar pressure distribution signals at the feature level to further improve the accuracy of FOG monitoring;Finally,using the fusion scheme of IMU and force-sensitive insoles,the IMU and force-sensitive insoles were integrated and optimized,and a prototype system of FOG monitoring and early warning was constructed and realized.The main work of this dissertation are as follows:(1)Aiming at the poor robustness of the existing FOG detection methods based on IMU signals,a new FOG detection method was designed according to the characteristics of IMU signals.The method adopts Convolutional Neural Networks(CNN)to automatically learn feature representation and Recurrent Neural Networks(RNN)to model the temporal dependencies between features.Squeeze and excitation blocks and attention mechanisms were introduced to improve the robustness of the model.The experimental results showed that the method achieved better detection results compared with the existing FOG detection methods based on IMU signals.Aiming at the cumbersome wearing process of IMU,through experimental comparison and analysis,it was determined that the best wearing position of IMU is bilateral ankles and the effective sampling frequency is above 30Hz.(2)Aiming at the problem that the research on the FOG detection method based on the plantar pressure distribution signals is still in its infancy,and the data scale is huge,a new FOG detection method was proposed according to the characteristics of the plantar pressure distribution signals.Firstly,an improved plantar pressure automatic segmentation algorithm was proposed,and a FOG detection model was constructed based on the time-frequency characteristics of the plantar pressure distribution signal,which verified that the plantar pressure segmentation is an effective method to extract the signal feature of the plantar pressure distribution with high robustness.Secondly,a Spatial-Stream-ConvNet,a Temporal-Stream-ConvNet,a Merge-Stream-ConvNet and a Three-Stream-ConvNet were constructed based on CNN to perform the FOG detection task.Compared with the other three network models,the Merge-StreamConvNet which utilizes the spatiotemporal map of plantar pressure constructed based on the plantar pressure segmentation algorithm and the periodic characteristics of gait realizes the data reduction of force-sensitive insoles and has higher robustness.(3)For highly robust FOG detection,a multi-modal fusion method for FOG detection was proposed.Based on the research results of the above two FOG detection methods,the multi-modal fusion method adopts an adaptive weighting method to perform feature-level fusion of the plantar pressure distribution signals and the IMU signals.Compared with the single-modal model,the multi-modal fusion method improves the accuracy of FOG detection.In addition,a pre-freezing category automatic labeling method based on freezing index ratio was proposed,and a multi-modal fusion model was used for FOG prediction.Compared with directly labeling the data 2.5 seconds before FOG onset as the pre-freezing category,the automatic labeling method generates a larger number of pre-freezing samples and has better prediction accuracy.(4)Based on the multi-modal fusion FOG monitoring method proposed in this paper,the IMU and the force-sensitive insoles were integrated and optimized to construct and implement a FOG monitoring and early warning prototype system.The system can meet the needs of patients for real-time monitoring and early warning of FOG,and also meet the needs of doctors for auxiliary diagnosis and management of freezing symptoms.