| Long-term exposure to pesticides,especially herbicides and insecticides,increases the risk of movement disorder,but the underlying mechanisms are currently unclear.Since the neurophysiological changes of the brain induced by pesticides occur earlier than the occurrence of movement disorders,and the slight motor symptoms in the early stage of the disease are not easy to detect,how to conduct effective quantitative assessment in neurophysiology and motor behavior for the delayed and progressive lesions induced by pesticides becomes the key to disease warning and prevention.Therefore,this paper proposes a novel experimental paradigm of pesticide-induced movement disorder in rats.Combined behavioral and neuroelectrophysiological techniques,the evolution relationship between pesticide exposure and movement disorder is explored,and biomarkers that can effectively assess the development of movement disorder are found.Then aiming at high-risk groups exposed to pesticides,an intelligent assessment model of movement disorders is constructed based on wearable sensors and intelligent detection algorithms.Models are interpreted visually to improve confidence,providing a powerful tool for pesticide safety protection and early warning of diseases in agricultural production.The specific work of this paper is as follows:(1)A novel experimental paradigm for pesticide-induced movement disorder in rats is proposed.Personalized electrode is implanted in the V layer of the primary motor cortex of the rat brain,and a micro-osmotic pump with continuous drug delivery for 28 days is implanted subcutaneously between the scapulae.It eliminates the need to frequently handle animals,overcomes the motor behavior and neurophysiological changes caused by acute toxins in traditional modeling.It accurately simulates the absorption mode of pesticides into human body through the skin in the process of agricultural production,and reproduced the disease development of human beings in the process of long-term exposure to low-doses pesticides to the greatest extent.Moreover,it provides experimental conditions for further research on the relationship between pesticide exposure and movement disorder and for finding biomarkers that can effectively assess the occurrence and development of movement disorder.(2)Combining behavioral and neuroelectrophysiological techniques,the evolutionary relationship between pesticide exposure and movement disorders is explored.Firstly,the effect of pesticide on the dopaminergic pathway in the nigrostriatum of rats is verified from the viewpoint of pathology and molecular biology,which confirms the validity of the experimental paradigm.The results of behavioral tests show that,excluding the effect of surgical trauma,the motor ability of rats is less affected by pesticide in the first week,the degree of motor impairment increase sharply in the second and third weeks,and slightly slowdown in the fourth week.After four weeks of administration,the total distance of movement is reduced by about56%,and the frequency of forelimb use is reduced by about 52%.At the same time,longitudinal tracking and analysis of neuroelectrophysiological data show that after continuous exposure to pesticides for 4 weeks,the power in beta band(13-35 Hz)and gamma band(35-80 Hz)increase by about 44% and 69%,respectively,and the average firing rate of cortical pyramidal neurons decrease by about 34%.It provides an effective biomarker for assessing the movement disorder induced by pesticide exposure,and provides a basis for screening patients with movement disorder for the construction of the movement disorder assessment model in the next step.(3)Based on wearable sensors and improved random forest algorithm,an intelligent assessment model is constructed for pesticide expose-induced movement disorders.Firstly,the EEG of 70 patients with movement disorders who had a long history of pesticide exposure is collected by portable electroencephalograph,and 50 patients with movement disorders with significant increases in beta and gamma frequency oscillations are selected based on biomarkers.Then,the movement data are collected from 50 patients and 50 healthy subjects during a 10-meter straight walk test using a body sensor network consisting of five inertial measurement units worn on the wrist,ankle and waist.After preprocessing the original data,the time domain and frequency domain features are extracted manually.The feature selection method based on tree is used for feature selection,and the random forest model is trained separately for each sensor data.Then the genetic algorithm is used to further optimize the model parameters,and the model performance is significantly improved.The results show that only one sensor placed at the waist position was sufficient to describe the movement status of the subjects(AUC =0.9706 ± 0.0016),which can effectively distinguish the patients with movement disorders from the healthy subjects with an accuracy of 90.5% ± 0.76%.Compared with other classifiers,it also has the best classification performance.Finally,SHAP method is used to explore the contribution of each feature to the model output,explaining the reason why the waist sensor driven random forest model has the best classification performance.(4)Considering the limitations of machine learning algorithms,an intelligent assessment model is reconstructed based on Convolutional Neural Network(CNN).Firstly,the preprocessed data are processed by continuous wavelet transform to obtain more detailed timefrequency information.Based on the constructed 6-channel CNN,a model was trained separately for each location sensor.The results show that the CNN model driven by waist sensor data has the best performance(AUC = 0.9981 ± 0.0017)compared with other sensors.Waist is the best monitoring position for assessing the daily walking of subjects,and can accurately identify the movement disorders during walking with an accuracy of 98.01% ± 0.85%.Compared with traditional machine learning algorithm model,the performance is significantly improved.Then,the gradient weighted class activation mapping method is used to visually explain the constructed CNN model,exploring the decision-making mechanism behind the model,demonstrating the reasons behind the optimal waist sensor performance.Furthermore,the CNN model driven by waist sensor with the best performance is reconstructed into 3D CNN,and the contribution of each time series component to the model is explored combined with the visualization interpretation method.Based on the visualization results,the model is optimized,the redundant data with low contribution to the model is removed,and the new CNN model is retrained.The cost of data acquisition and processing was reduced by 50% while the performance is guaranteed(AUC = 0.9929 ± 0.0019),which provides the possibility for the development of daily gait real-time monitoring system.Finally,an easy-to-use graphical user interface is developed based on the optimal model and clinically deployed.In summary,this paper provides an ideal animal model for the exploration of the potential mechanism and prevention of pesticide exposure-induced movement disorder,ascertains the evolution relationship between pesticide exposure and movement disorder,and finds out biomarkers that can effectively assess pesticide exposure-induced movement disorder.In addition,a reliable movement disorder assessment model is developed for the high-risk groups of pesticide exposure engaged in agricultural production,and the optimal wearing position of the sensor is determined,which can effectively and quantitatively assess the movement behavior of the high-risk groups of pesticide exposure,providing a powerful tool for pesticide safety protection and disease early warning in agricultural production. |
