| BackgroundLumbar disc herniation(LDH)is one of the major causes of low back pain and lower limb neurogia.With its high incidence,high recurrence rate,long course of disease and low age trend,LDH affects human health and life quality seriously.The highest incidence of LDH was in L4/L5 and L5/S1 and causes corresponding nerve root compression.Currently,clinical symptoms,physical examination and imaging results are usually combined to confirm the diagnosis of LDH.However,the disc herniation does not always correspond to the distal nerve root segment to which the neuroglia symptoms are directed.Straight Leg Raising Test(SLRT)is one of the most commonly used tests for the diagnosis of nerve root compression.The sensitivity of SLRT is high,but the specificity is insufficient,which makes its ability to prediction of nerve root remains unclear.Magnetic resonance imaging(MRI)is the gold standard to characterise the morphology of intervertebral discs,assess the surrounding soft tissues and identify whether nerves were involved,which has a diagnostic specificity of up to 97%or more in LDH diagnosis.However,some people without neurological symptoms may also have radiographic signs of disc herniation,suggesting that independent radiographic evidence may not be completely consistent with the specific functional limitations of LDH patients.MRI examinations are mostly completed in the resting position,which cannot fully reflect the physiological function of the lumbar spine,and also lacks the dynamic monitoring and evaluation of nerve root compression under movement state and the changes in neuromuscular function caused by it.Therefore,it is necessary to explore whether there are other dynamic assessment methods that can provide useful supplement to the existing LDH localization diagnosis from the perspective of function,which is also one of the clinical challenges rehabilitation physicians are facing.Gait analysis and surface electromography(SEMG)are commonly used for functional assessment,which can quantitatively evaluate individual's gait characteristics and dynamic neuromuscular function.LDH patients,especially those with lower limb neuroglia,often present with an adaptive gait changes such as reduced time to touch the ground,slower frequency and speed on the symptomatic side.The neuromuscular control of lumbar region and lower limbs is also adjusted accordingly,showing a decrease in muscle strength and endurance.However,it is still unclear whether there are differences in gait characteristics and muscle activation patterns among patients with different nerve root compression,and whether this difference can provide function basis for nerve root localization diagnosis.In addition,there are too many parameters in gait analysis and SEMG,such as the spatiotemporal,kinematics and dynamics parameters and the time/frequency domain parameters about muscle activation,coordination and fatigue.So it is also worth further exploration how to extract highly sensitive indicators that are helpful to distinguish different nerve root compression.As a highly flexible processing method,machine learning technology can excavate valuable data through learning and training from a large number of data and construct effective prediction models.In recent years,machine learning has been gradually applied in the medical field,which provides the possibility for the classification,selection and extraction of gait and SEMG parameters and assisting clinical decision making.Accordingly,in this study,wearable gait analysis and SEMG detection were used to synchronize the gait and SEMG analysis of LDH patients with different nerve root compression,and the gait characteristics and neuromuscular activation regularity were summarized.Random Forest(RF)and Back Propagation(BP)neural network were selected to establish the practical and available localization diagnosis model based on gait and SEMG parameters,which provides a functional basis and supplement for LDH localization diagnosis,and also provided a new idea for the in-depth application of evaluation methods for LDHObjectiveThe first aim of this study was to summarize the gait characteristics and muscle activation patterns of patients with different nerve root compression;The second one was to analyze the effectiveness of the diagnosis model based on gait and SEMG parameters;And the third one was to explore the application value of gait and SEMG characteristics in LDH localization diagnosisSubjects and Methods1.Subjects:In this study,58 LDH patients with radiating pain in either lower extremity who underwent surgical treatment were recruited,among which 29 patients have compression on L5 nerve root(L5 group)and 29 patients on S1 nerve root(S1 group).Meanwhile,30 healthy subjects were recruited as healthy control group.2.Gait analysis and SEMG detection:IDDEA gait analyzer was used to collect the spatiotemporal and kinematic parameters of the subjects;DELSYS wireless surface EMG test system and GoPro Hero3 high-speed camera were used to collect the electrical activity of tibialis anterior(TA)and lateral gastrocnemius(LG)of lower extremities during walking.During the test,participants were instructed to walk at their comfortable speed,back and forth on a 10-meter track and at least 20 complete gait cycles were collected for each participant.The gait spatiotemporal parameters such as speed,cadence,step length,stride length,single leg support time,double leg support time,single/double support time ratio,step duration and cycle duration,the four acceleration parameters(pulling acceleration,swing power,ground impact,foot fall control)and push-off angle in pre-swing stage were calculated by ActView3 software.Then the gait characteristics of LDH patients during walking were analyzed and summarized.Analyzing and summarizing the muscle activation characteristics of LDH patients was through MATLAB program to analyze 20 gait cycles,calculate the peak root mean square(RMS-peak),RMS-peak time(the position of RMS-peak in a gait cycle),mean power frequency(MPF),median frequency(MF)of TA and LG,which was based on the EMG data preprocessed by EMGwork Analysis software.3.Establishment of diagnosis model:The BP neural network and RF algorithm were applied based on SEMG parameters,gait parameters and combined parameters of gait and SEMG after screening of characteristic between bilateral lower limbs,six different diagnosis models were established and verified according to the principle of repeated reservation experiment to calculating the accuracy,precision,recall rate,F1-score and Kappa value of each experiment and evalating the area under the ROC curve were also used to evaluate the diagnostic efficiency of the six diagnosis models.Then an auxiliary localization diagnosis system of compressed nerve root was developed.Results1.Characteristics of gait and SEMG(1)Spatiotemporal characteristicsLDH patients showed decreased walking stability and symmetry.Compared with the asymptomatic side,the speed(P=0.001),cadence(P<0.001),single leg support time(P=0.002)and step duration(P=0.011)in the symptomatic side of L5 group were significantly decreased.The cadence(P=0.004),single leg support time(P<0.001),double leg support time(P=0.020)in the symptomatic side were significantly decreased compared with the asymptomatic side in S1 group.Compared with healthy control group,the speed(P=0.022,P<0.001),cadence(P=0.034,P=0.009),step length(P=0.002,P<0.001),stride length(P=0.002,P<0.001),single/double support time ratio(P<0.001,P<0.001)were significantly decreased,double leg support time(P=0.01,P=0.001)was significantly increased in L5 and S1 group.In addition,step duration(P=0.005)and cycle duration(P=0.033)were significantly increased in S1 group(2)Kinetics characteristicsThe swing power(P=0.005,P=0.002),ground impact(P<0.001,P=0.001),foot fall control(P<0.001,P=0.001)and push-off angle(P=0.007,P=0.004)in the symptomatic side were significantly lower than those in asymptomatic side in L5 and S1 groups.Compared with healthy control group,the pulling acceleration(P=0.018)and foot fall control(P=0.016)in L5 group were significantly decreased,and the pulling acceleration(P=0.008),swing power(P=0.035),ground impact(P=0.008),foot fall control(P=0.01)in S1 group were significantly decreased.(3)SEMG characteristicsIn L5 Group,compared with the asymptomatic side,the RMS-peak time of TA in the symptomatic side was significantly delayed(P<0.001),the MPF and MF were significantly decreased(both P<0.001).At the same time,the RMS-peak of LG in the symptomatic side was significantly decreased(P=0.016).In S1 group,compared with the asymptomatic side,as for LG,the RMS-peak in the symptomatic side was significantly decreased(P=0.003),the RMS-peak time was significantly forward(P<0.001),the MPF and MF were significantly decreased(both P<0.001),and the RMS-peak of TA in the symptomatic side was also significantly decreased(P=0.043).When compared with the healthy control group,in L5 group,the RMS-peak time of TA was significantly delayed(P<0.001),the MF of TA(P=0.002)and MPF of LG(P=0.001)were significantly decreased.In S1 group,the RMS-peak(P=0.043)and MPF(P<0.001)of LG,MPF(P=0.033)and MF(P=0.001)of TA were significantly decreased.In addition,when compared with S1 group,the RMS-peak time of TA in L5 group was significantly delayed(P<0.001),and the RMS-peak time of LG was significantly decreased(P=0.001).In contrast,when Compared with the L5 group,the RMS-peak time of LG in S1 group was significantly forward(P=0.045).(4)Lower limb muscle activation during walkingWhen L5 nerve root was compressed,the activation time of anterior tibial was significantly delayed,which was manifested by a later occurrence in the gait cycle[symptomatic side:35(25.5,61),asymptomatic side:11(7.5,15),P<0.001]and resulted in co-contraction with LG.When S1 nerve root was compressed,the activation of LG moved forward significantly,which was manifested by an earlier occurrence of RMS-peak in the gait cycle[symptomatic side:27(14.5,37.5),asymptomatic side:44(38.5,46.5),P<0.001],showing a bimorphic activation pattern and a tendency of co-contraction with TA.2.Establishment of diagnosis models and development of diagnostis system(1)Establishment and comparison of diagnosis modelsSix diagnosis models were established using RF and BP neural network based on eight gait parameters,eight SEMG parameters and eight combined parameters of gait and SEMG.The results showed that the average accuracy of ten diagnoses of three models using BP neural network was 77%,75%and 81%,the highest accuracy was 84%,86%and 91%,the Kappa values were 0.68,0.67,0.73,and the areas under the ROC curve were 0.89,0.88,0.93,respectively.The average accuracy of ten diagnoses of the three models using RF algorithm is 80%,84%,89%,the highest accuracy is 86%,93%,95%,the Kappa value is 0.73,0.76,0.85,and the area under the ROC curve is 0.90,0.93,0.97,respectively.It is suggested that the RF diagnosis model based on the combined parameters of gait and SEMG has the best diagnostic efficiency.(2)Weight comparison of different diagnostic modelsAmong three RF models,the weight of eight parameters in the gait model was nearly equal(10%-15%);the weight of eight parameters in the SEMG model ranged from 6%-26%,of which the RMS-peak time of TA was the highest(26%),followed by the RMS-peak and RMS-peak time of LG(15%and 15%,respectively).In the combined gait and SEMG model,the weight of eight parameters ranged from 7%-27%,of which the RMS-peak time of TA accounted for the highest(27%),followed by the RMS-peak of TA,MPF of LG and ground impact(14%,13%,13%,respectively)(3)Development of diagnosis systemWe developed an auxiliary localization diagnosis system on the basis of diagnosis models,designed user interface which is suitable for intelligent medical diagnosis system,established communication between users and system through the visual interactive button,which improved usability of the diagnosis system obviously.The system can store basic information,output the evaluation indicators of models such as accuracy in real time and also has 6 built-in diagnostis models to be freely selected for diagnosis.Conclusions1.LDH patients with L5 and S1 nerve root compression showed similar decreased walking stability and bilateral asymmetry,presenting with abnormal gait spatiotemporal and kinematic characteristics.The limitation of ankle-plantar flexor function was more significant in patients with S1 nerve root compression,which affected forward propulsion during walking,and the decrease of walking ability was more obvious.2.LDH patients with L5 and S1 nerve root compression showed distinct patterns of lower limb neuromuscular activation.When the L5 nerve root was compressed,the nerve control disorder occurred in TA,and the activation time of TA was significantly delayed in the gait cycle.Dysfunction of gastrocnemius muscle during S1 nerve root compression was characterized by bimodal activation and earlier activation.Such changes in the muscle activation pattern led to the co-contraction and coordination disorder between the antagonistic muscles.3.Based on gait and SEMG characteristics,the diagnosis models using BP neural network and RF algorithms can be used as auxiliary method for compressed nerve roots,providing reference for accurate diagnosis of LDH,among which the RF model based on combined gait and SEMG parameters has the best diagnostic efficiency.4.As for the localization of L5 and S1 nerve roots in LDH patients,SEMG characteristics are more advantageous and valuable than spatiotemporal and kinematic characteristics,among which the RMS-peak time of TA is the most sensitive diagnostic index. |
PDF Full Text Request