Quantitative Analysis Of Speech In Parkinson’s Disease Patients With Freezing Of Gait

Background and objective: Freezing of gait(FOG)is a common disabled gait abnormality in patients with Parkinson’s disease(PD).It is the main reason for the increase of fall risk,the decrease of functional independence and the decline of quality of life.Because of the complex pathological mechanism and varied clinical manifestations of FOG,it is difficult to diagnose and evaluate it in clinical practice.Both speech disorders and FOG are considered to be axial symptoms of PD,and they may have a common pathological mechanism.Some studies have reported the correlation between speech parameters and axial motor symptoms such as posture and gait.Quantitative analysis of speech in PD patients with FOG is helpful to further explore the association between speech features and FOG and may be applied to the diagnosis and prediction of FOG in PD patients.Therefore,this study firstly analyzed the voice handicap index(VHI)and clinical features of PD patients with and without FOG to explore the related factors of FOG in PD patients and the relationship between voice impairment and FOG.Then the speech parameters of PD patients with and without FOG were analyzed by speech quantitative analysis technique,to evaluate the effect of levodopa on the speech parameters of PD,and to further explore whether there are speech phenotypic differences between PD with and without FOG.Finally,based on the "On state" speech or gait parameters,the machine learning algorithm was used to build classification models to explore the feasibility of "On state" speech or gait features in recognition of FOG in PD patients.Methods:This study was divided into three parts.The first part included PD patients who visited the Parkinson’s Disease outpatient department of neurology,The First Affiliated Hospital of Chongqing Medical University from July 2019 to July 2021.Demographic and clinical data of all patients were collected.All patients underwent neuropsychological tests,including cognition,sleep,anxiety,depression.The Movement Disorder Society-Unified Parkinson’s Disease Rating Scale(MDS-UPDRS)and H&Y stage were used to evaluate disease severity.VHI and Freezing of Gait Questionnaire(FOGQ)were used to assess voice impairment and FOG of PD.According to FOGQ,patients were divided into two groups,PD with FOG(PD-FOG)and PD without FOG(PD-n FOG).The demographic characteristics,clinical characteristics and VHI were compared between the two groups.Multivariate Logistics regression analysis was used to explore the related factors of FOG and the relationship between voice impairment and FOG.In the second part,PD patients with and without FOG,and age-and gender-matched healthy controls(HC)were recruited.The demographic and clinical data of all participants were collected.All patients underwent speech and motor symptoms tests during medication "off" and "on" state,respectively.The speech task was sustained vowel phonation.Voice parameters were extracted by the Praat software.Paired t-test or Wilcoxon signed rank test were used to detect the difference of voice parameters between "off" and "on" state.Spearman correlation analysis was used to explore the correlation between the changes of speech parameters and the improvement of motor symptoms.In the third part,PD patients with dopamine-responsive FOG(PD-FOG)and without FOG(PD-n FOG)were recruited in this study.Demographic and clinical data of all patients were collected.All patients underwent MDS-UPDRS,H&Y and neuropsychological tests.Gait detection and voice sample collection were completed during the "On state" of patients(1-2 hours after medication).Gait detection included normal walking,fast walking and dual task walking.The speech task was sustained vowel phonation.A portable gait analyzer based on multi-sensor was used to collect gait information and analyze gait parameters.Praat software was used to extract speech features.The clinical characteristics,speech and gait parameters were compared between the two groups.The correlation between speech and gait parameters and clinical characteristic was analyzed by Spearman correlation analysis.Based on speech and gait features,five classifiers including K-nearest neighbor(KNN),support vector machine(SVM),decision tree,random forest(RF)and Adaboost were used to classify PD-FOG and PD-n FOG in a supervised machine learning setup.The genetic algorithm was used to select features and analyze the importance of each feature.Results:1.In the first part of this study,a total of 641 PD patients were enrolled,including 202 PD-FOG and 439 PD-n FOG.By comparing the demographic data and clinical characteristics of patients in the two groups,we found that 1)compared with PD-n FOG,the patients in the PD-FOG group were older,and had longer disease duration,higher H&Y stage,and higher daily equivalent dose of levodopa(LED).The MDS-UPDRS Ⅰ、Ⅱ、Ⅲ、Ⅳ scores in PD-FOG group were higher than those in PD-n FOG group.The Mini-Mental State Examination(MMSE)and Parkinson’s disease Sleep Scale(PDSS)scores were lower in the PD-FOG group than in the PD-n FOG group,and Epworth Sleepiness Scale(ESS),Rapid Eye Movement Sleep Behavior Disorder Questionnaire(RBDQ),Hamilton Depression Score(HAMD)and Hamilton Anxiety score(HAMA)were higher than those in PD-n FOG group.2)After adjusting for age,disease duration,H&Y stage and MDS-UPDRS Ⅲ scores,the scores of VHI in PD-FOG group were significantly higher than those in PD-n FOG group.3)Multivariate Logistics regression suggested higher VHI scores(OR=1.022,P=0.001),H&Y stage(OR=2.793,P<0.001),MDS-UPDRS III scores(OR=1.026,P=0.002)and LED(OR=1.002,P<0.001)were independent factors for FOG in patients with PD.2.In the second part of this study,we further analyzed the speech parameters of 30 PD-FOG,30 PD-n FOG and 30 HC during "on" and "off" state and found that 1)after adjusting for age,gender and MDS-UPDRS Ⅲscores,the local amplitude perturbation(Shimmer),three-point amplitude perturbation quotient(APQ3),the average absolute difference between consecutive differences between the amplitudes of consecutive periods(DDA)and noise harmonic ratio(NHR)were higher in PD-FOG group than in PD-n FOG group.2)The scores of rigidity,bradykinesia and tremor in both PD-FOG and PD-n FOG groups were improved after Levodopa(L-dopa)intake.In addition,the score of postural instability/gait disorder in PD-FOG group was also improved.There was no significant improvement in the scores of speech and facial expression after medication in the two groups.3)There was no significant improvement in 27 speech features of PD-FOG patients after medication.The local jitter,absolute local jitter(Jitta),relative amplitude perturbation(RAP),five-point period perturbation quotient(PPQ5),the ratio of average absolute difference of differences between cycles to the average period(DDP)and the noise to harmonic ratio(NHR)decreased,while the MPT increased in PD-n FOG patients after L-dopa intake.In PD-n FOG group,the changes of NHR were correlated with the improvement of MDS-UPDRS Ⅲ scores.3.In the third part of this study,we analyzed the speech and gait parameters of 63 PD patients with dopamine-responsive FOG and 87 PD without FOG during the "On state" and found that: 1)After adjusting for age,gender,disease duration,H&Y stage and MDS-UPDRS Ⅲ scores,the Jitter,Jitta,RAP,PPQ5,DDP,Shd B and the standard deviation of pulse period in PD-FOG group were higher than those in PD-n FOG group,while the harmonic to noise ratio(HNR),mean F0 and MPT were lower than those in PD-n FOG group.2)The Jitter(r=0.325),Jitta(r=0.237),HNR(r=-0.247),F0 SD(r=0.186)and MPT(r=-0.28)were all correlated with the disease duration of PD.The Jitter was correlated with the scores of facial expression(r=0.261),rigidity(r=0.181),bradykinesia(r=0.167)and postural instability/gait disorder(r=0.227).The mean F0 was negatively correlated with the scores of speech disorder(r=-0.243).The F0 SD was positively correlated with the scores of bradykinesia(r=0.235)and postural instability/gait disorder(r=0.231).The pulse period SD was also positively correlated with the scores of bradykinesia(r=0.224)and postural instability/gait disorder(r=0.233).The MPT was negatively correlated with the scores of speech(r=-0.196),facial expression(r=-0.169),rigidity(r=-0.234),bradykinesia(r=-0.272)and postural instability/gait disorder(r=-0.394).There was no correlation between all voice parameters and tremor scores.3)Based on the "On state" speech features,the accuracy of five classifiers in classifying PD-FOG and PD-n FOG was range from75.95% to 85.95%,the sensitivity was range from 79.05% to 88.57%,and the specificity was range from 63.33% to 85.71%.In 10 cross-validation experiments,the most frequently selected speech features were HNR,min F0,pulse period and MPT.4)After adjusting for age,gender and height,compared with PD-n FOG,PD-FOG had shorter stride time and step time,larger cadence,larger variation coefficient of stride time and cadence during three walking tasks.The pulling acceleration,swing power and ground impact of PD-FOG were higher than those of PD-n FOG,while pre-swing angle of PD-n FOG was smaller than that of PD-n FOG.5)Five classifiers were used for classifying PD-FOG and PD-n FOG based on gait parameters of three walking tasks,and gait parameters of normal walking had the best performance for classification(the average accuracy ranged from 78.33% to 89.05%,the average sensitivity ranged from 80% to90.95%,and the average specificity ranged from 76.19% to 87.14%).Among 10 cross-validation experiments,the most frequently selected gait parameters were the stance phase(%),the pulling acceleration and the pre-swing angle,followed by step length,velocity,and ground impact.6)Combining speech and gait features for classification,the accuracy of all classifiers except SVM were improved,and the Adaboost had the highest accuracy,reaching 93.10%.Conclusion:1.Voice impairment is an independent factor for FOG in PD.Speech analysis of PD patients may be helpful in FOG assessment and prediction.2.Dopamine replacement therapy can improve part of speech parameters in PD-n FOG,but not in PD-FOG,suggesting that there may be speech phenotypes differences between PD-FOG and PD-n FOG.3.PD patients with dopamine-responsive FOG showed more severe voice and gait impairment patterns than PD-n FOG during "On state".The machine learning algorithm based on " On state " speech and gait features had well performance in detecting PD with "Off state" FOG.Speech and gait analysis is expected be used as a supplement tool for FOGQ to identify and evaluate FOG in PD.