Design Of Vestibular Function Detection System Based On Inertial Sensors

Vertigo caused by vestibular dysfunction is the third major clinical symptom,which seriously affects people’s normal life.The clinical manifestation of vertigo is unsteadiness of standing,spontaneous dumping,nausea,vomiting and so on.At present,the clinical commonly used detection methods include video-head impulse test,dynamic visual acuity test,electronystagmography,caloric test and so on.These methods are complex,time-consuming and can only detect static vestibular function,at the same time in the process of testing in patients with easy to produce the unwell symptom such as dizziness,nausea,vomiting,increased in patients with pain.On the other hand,the installation of related testing instruments requires a large space,and it is not easy to move after fixed installation.Moreover,the maintenance cost are high,which increases the medical burden of patients.Based on modern science and technology,it is of great clinical significance and social value to explore a simple,rapid,economical and accurate new method for dynamic vestibular function detection and to construct a vestibular function detection model for clinical computer-assisted detection.The clinical manifestation of vestibular dysfunction is vertigo and imbalance.Posture signal can reflect the balance state of human body during walking,and then reflect the vestibular function of human body.In this paper,inertial sensors were introduced into vestibular function detection.Based on the 8 walking styles in the dynamic gait index to obtain human posture signals.And feature matrix was established by feature extraction,the optimal feature subset was obtained by feature evaluation.The vestibular function recognition model was constructed by integrated learning method.Finally,the vestibular function auxiliary detection platform was designed to realize the real-time detection of vestibular function.The main research contents of this paper are as follows:(1)Human posture signal acquisition based on inertial sensor.A data acquisition scheme based on signal acquisition terminal is designed,in which the signal acquisition terminal is composed of inertial sensor and bluetooth module.Posture signals were collected from 45 patients with vestibular dysfunction and 30 healthy controls under 8 walking modes.Among them,the inertial sensor is composed of triaxial accelerometer,triaxial gyroscope and triaxial magnetometer.And the 8 kinds of walking styles in the dynamic gait index including walking at normal speed,walking at changing speed,walking with horizontal head rotation,walking with vertical head rotation,turning and stopping while walking,walking over obstacles,walking around obstacles and walking up and down stairs,to get as much attitude information as possible.(2)Human posture signal processing and feature subset construction.After unit conversion and signal processing,the noise was removed by least square smoothing filtering.Based on the preprocessed signals,time domain,frequency domain and nonlinear characteristics were extracted,among which time domain features include mean value,variance,root mean square,skewness and kurtosis,frequency domain features include DC component and signal energy,and nonlinear features include sample entropy and fuzzy entropy.Then,the Relief algorithm was used to evaluate the feature set,and the importance of each feature is obtained.According to the weight of different features,the feature was assigned and the optimal feature subset was obtained.(3)Construction of vestibular function recognition model.A vestibular function recognition model based on ensemble learning was constructed by using grid optimization and tenfold cross optimization algorithms.In this model,support vector machine and random forest were selected as the meta-classifiers,and then support vector machine fusion meta-classifiers were used to identify vestibular functions based on the combined signal features.(4)Design and implementation of auxiliary detection platform.In this paper,a vestibular function assisted detection platform based on inertial sensors was designed.The platform consists of login module,data receiving module,signal real-time display module,detection result and report output module and other functional modules.After testing,the platform can preliminarily realize the real-time and accurate detection of vestibular function.After optimization,it can be used for clinical auxiliary diagnosis of vestibular function.