Computer-assisted Speech Analysis In Patients With Mild Cognitive Impairment And Alzheimer’s Disease

Part One Reliability and validity analysis of the Chinese version of the Addenbrooke’s cognitive examination-Ⅲ Background: With an aging population,the prevalence,morbidity and disability rate of Alzheimer’s disease(AD)are increasing gradually,which causes serious social problems and heavy burden for families.Therefore,it is very important to identify the early-stage of AD and its precondition—mild cognitive impairment(MCI)for treatment and intervention.Due to the lack of effective neuropsychological tests to identify MCI and mild AD in clinical practices so-far,it is urgent to develop appropriate neuropsychological screening tests for it.The Addenbrooke’s cognitive examination-Ⅲ(ACE-Ⅲ)was revised in 2012 and recommended as a screening tool for early-stage of AD,with a good sensitivity and specificity in previous studies.Unfortunately,the reliability and validity of the Chinese version of ACE-Ⅲ has not been reported yet.Objective: To import,translate and revise the ACE-Ⅲ for producing the Chinese version of ACE-Ⅲ,and then analyze the reliability and validity of Chinese version of ACE-Ⅲ,calculate the optimal cut-offs to detect MCI and AD and compare it with MMSE.Methods: In the present study,27 probable AD patients,25 MCI patients and 31 normal control subjects(NC)were recruited.All the participants completed the neuropsychological tests of Chinese version of ACE-Ⅲ,MMSE and Clinical Dementia Rating Scales(CDR).The internal consistency reliability,split-half reliability,content validity,structure validity and criterion validity of Chinese version of ACE-Ⅲ and its value of diagnosis differenation among the three groups were analyzed.The ROC curves of detecting MCI and AD by Chinese version of ACE-Ⅲ were made,the optimal cut-offs were calculated and the accuracy of Chinese version of ACE-Ⅲ and MMSE was compared.Results: The Chinese version of ACE-Ⅲ had good internal consistency reliability(Cronbach’s Alpha coefficient was 0.910)and split-half reliability(Spearman-Brown coefficient was 0.907).Its correlation with MMSE was good(Pearson coefficient was 0.957,P<0.001).To detect MCI,the area under the curve(AUC)of Chinese version of ACE-Ⅲ was 0.926,larger than the AUC value of 0.837 of MMSE,the optimal cut-off to detect MCI was 85/86,with sensitivity of 88.0% and specificity of 90.0%,which was superior to MMSE with sensitivity of 80.0% and specificity of 70.0% at the optimal cut-off of 27/28.To detect AD,the AUC of Chinese version of ACE-Ⅲ was 0.981,smaller than the AUC value of 0.990 of MMSE,the optimal cut-off to detect AD was 63/64,with sensitivity of 96.2% and specificity of 94.5%,which was not superior to MMSE with sensitivity of 100.0% and specificity of 85.5% at the optimal cut-off of 23/24.Conclusion: The present study has confirmed that the Chinese version of ACE-Ⅲ has a good reliability and validity,with a good accuracy to detect MCI,which is superior to MMSE.The Chinese version of ACE-Ⅲ is recommended as a screening test to detect MCI and early-stage of AD.Part Two Computer-assisted speech analysis in patients with mild cognitive impairment and Alzheimer’s disease Background: Cognitive impairment in Alzheimer’s disease(AD)involves multiple cognitive domains.In addition to learning and memory,language dysfunction has been reported closely related to the progression of AD.There is slight but definite language dysfuncction at the prodromal stage of AD or MCI.Computer-assisted speech analysis has already been applied in clinical evaluations on other diseases such as dysathrias and Parkinson’s disease(PD).The clinical evaluations on language function are still limited to traditional pen and paper tests.Our research group independently developed the "ASR speech recognition software for cognitive impairment screening V1.3".Mandarin vowels reading samples and spontaneous continuous speech samples more representative of the language performance of daily life can be analyzed objectively,accurately and quickly by the software,generating voice acoustic parameters and continuous speech parameters,which is prospective for clinical application.Objective: To evaluate the language characteristics in patients with MCI and early-stage AD assisted with the "ASR speech recognition software for cognitive impairment screening V1.3".Methods: 20 AD patients,20 MCI patients and 24 healthy elderly controls were recruited from Shanghai Ruijin Hosptial,respectively.Each subject underwent the neuropsychologic tests of mini-mental state examination(MMSE),Addenbrooke’s cognitive examination-Ⅲ(ACE-Ⅲ)and Clinical Dementia Rating Scales(CDR).Spontaneous speech samples were recorded through the description task of Cookie-Theft Picture from Boston Diagnostic Aphasia Examination.Mandarin vowels reading samples were also obtained.The voice recordings were processed and analyzed by the computerized software to generate voice acoustic parameters and 24 continuous speech parameters.Analysis Of Variance(ANOVA)or non-parametric test was used to compare the differences of the speech variables and neuropsychologic test scores among subject groups.Pearson correlation and stepwise multiple linear regression were performed to further identify the correlation between the speech variables and the neuropsychologic results.Results: There were no significant differences in the acoustic parameters among three groups,indicating that the phonological function of MCI and AD patients was still preserved.14 of the total 24 continuous speech parameters were significantly different among three groups.The most representative parameters were described as following: Mean values of the variable of Percentage of Silience Duration(namely the ratio of total silience duration to the total speech duration)were 31.1%±10.7%,41.8%±11.6%,47.0%±11.8% respectively in the NC、MCI and AD group,P<0.001.Mean values of the variable of Average Duration of Silence Segment among groups were 0.9±0.4,1.2±0.4,1.6±0.8(s)respectively,P=0.008.Both of the variables were significantly different between NC and MCI group,as well as between NC and AD group by post hoc comparisons.Both variables were significantly negatively related with MMSE and ACEⅢ total score.Results of stepwise multiple linear regression showed that variable of Average Duration of Silence Segment was most closely related with MMSE and ACE-Ⅲ scores among the 24 prarmeters.We found that both of the two variables were helpful to identify MCI and early-stage of AD from the normal ederly,while Average Duration of Silence Segment showed to be a valuable indicator of cognitive impairment severity.Mean values of the variable of Count of Long Pause(namely silence segment with the duration ranges from 400 ms to 1000ms)were 6.1±3.5,6.4±2.8,3.5±1.8 respectively in the NC 、 MCI and AD group,P=0.008.Mean values of the variable of Count of Voice Segment among groups were 0.9±0.4,1.2±0.4,1.6±0.8 respectively,P=0.008.Both of the variables were significantly different between MCI and AD group by post hoc comparisons.While the variable of Long Pause/Voice Count Ratio and variable of Average Duration of Voice Segment showed no differences among three groups.The results implied that the amount of expressive language initially increased from NC to MCI stage,than decreased dramatically from MCI to AD stage,showing that there were different language characteritics at different stages of cognitive impairment.MCI patients,accompanied with mild difficulty in words finding,language planning and abstracting,had more talking with more pauses when expressing the same content with normal ederly,while AD patients showed more silence duration at each pause and few amount of talking.Variables of Count of Long Pause and Count of Voice Segment showed to be very important indicators to help distinguish MCI with AD.Conclusions: Computer-assisted speech analysis showed to be a noninvasive and a promising assessment tool in the evaluation and differential diagnosis of cognitive impairment.