Analysis of Smartphone Recordings in Time, Frequency, and Cepstral Domains to Classify Parkinson’s Disease
2
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
Objectives: Parkinson’s disease (PD) is the second most common neurodegenerative disorder; it affects more than 10 millionpeople worldwide. Detecting PD usually requires a professional assessment by an expert, and investigation of the voice as abiomarker of the disease could be effective in speeding up the diagnostic process. Methods: We present our methodology inwhich we distinguish PD patients from healthy controls (HC) using a large sample of 18,210 smartphone recordings. Thoserecordings were processed by an audio processing technique to create a final dataset of 80,594 instances and 138 featuresfrom the time, frequency, and cepstral domains. This dataset was preprocessed and normalized to create baseline machinelearningmodels using four classifiers, namely, linear support vector machine, K-nearest neighbor, random forest, and extremegradient boosting (XGBoost). We divided our dataset into training and held-out test sets. Then we used stratified5-fold cross-validation and four performance measures: accuracy, sensitivity, specificity, and F1-score to assess the performanceof the models. We applied two feature selection methods, analysis of variance (ANOVA) and least absolute shrinkageand selection operator (LASSO), to reduce the dimensionality of the dataset by selecting the best subset of features that maximizesthe performance of the classifiers. Results: LASSO outperformed ANOVA with almost the same number of features. With 33 features, XGBoost achieved a maximum accuracy of 95.31% on training data, and 95.78% by predicting unseen data. Conclusions: Developing a smartphone-based system that implements machine-learning techniques is an effective way todiagnose PD using the voice as a biomarker.
Abstract
Objectives: Parkinson’s disease (PD) is the second most common neurodegenerative disorder; it affects more than 10 millionpeople worldwide. Detecting PD usually requires a professional assessment by an expert, and investigation of the voice as abiomarker of the disease could be effective in speeding up the diagnostic process. Methods: We present our methodology inwhich we distinguish PD patients from healthy controls (HC) using a large sample of 18,210 smartphone recordings. Thoserecordings were processed by an audio processing technique to create a final dataset of 80,594 instances and 138 featuresfrom the time, frequency, and cepstral domains. This dataset was preprocessed and normalized to create baseline machinelearningmodels using four classifiers, namely, linear support vector machine, K-nearest neighbor, random forest, and extremegradient boosting (XGBoost). We divided our dataset into training and held-out test sets. Then we used stratified5-fold cross-validation and four performance measures: accuracy, sensitivity, specificity, and F1-score to assess the performanceof the models. We applied two feature selection methods, analysis of variance (ANOVA) and least absolute shrinkageand selection operator (LASSO), to reduce the dimensionality of the dataset by selecting the best subset of features that maximizesthe performance of the classifiers. Results: LASSO outperformed ANOVA with almost the same number of features. With 33 features, XGBoost achieved a maximum accuracy of 95.31% on training data, and 95.78% by predicting unseen data. Conclusions: Developing a smartphone-based system that implements machine-learning techniques is an effective way todiagnose PD using the voice as a biomarker.
참고문헌 (19)
- Parkinson J, "An essay on the shaking palsy" (1817), Sherwood, Neely and Jones
- Parkinson J, "An essay on the shaking palsy 1817", J Neuropsychiatry Clin Neurosci, 14(2), pp.223~236 (2002)
- Charcot JM, (1872), Aux bureaux du Progres Medical
- Charcot JM, "Lectures on the diseases of the nervous system:delivered at La Salpetriere" (1877), The New Sydenham Society
- Brissaud E, "Lecons sur les maladies nerveuses (Salpetriere, 1893-1894)" (1895), G. Masson
- Heisters D, "Parkinson’s : symptoms, treatments and research", Br J Nurs, 20(9), pp.548~554 (2011)
- Miller N, "Prevalence and pattern of perceived intelligibility changes in Parkinson’s disease", J Neurol Neurosurg Psychiatry, 78(11), pp.1188~1190 (2007)
- Ho AK, "Speech impairment in a large sample of patients with Parkinson’s disease", Behav Neurol, 11(3), pp.131~137 (1998)
- Tsanas A, "Accurate telemonitoring of Parkinson’s disease progression by noninvasive speech tests", IEEE Trans Biomed Eng, 57(4), pp.884~893 (2010)
- Little MA, "Suitability of dysphonia measurements for telemonitoring of Parkinson’s disease", IEEE Trans Biomed Eng, 56(4), pp.1015~ (2009)
- Benba A, "Voiceprints analysis using MFCC and SVM for detecting patients with Parkinson’s disease", Proceedings of 2015International conference on electrical and information technologies (ICEIT), pp.300~304 (2015)
- Hemmerling D, "Parkinson’s disease classification based on vowel sound", Proceedings of the 11th International Workshop on Models and Analysis of Vocal Emissions for Biomedical Applications (2019)
- "mPower: mobile Parkinson disease study" (2019)[URL]
- Bot BM, "The mPower study, Parkinson disease mobile data collected using ResearchKit", Sci Data, 3, pp.160011~ (2016)
- Giannakopoulos T, "Introduction to audio analysis: a MATLAB approach" (2014), Academic Press
- Giannakopoulos T, "pyAudioAnalysis: an open-source python library for audio signal analysis", PLoS One, 10(12), pp.e0144610~ (2015)
- Pedregosa F, "Scikit-learn : machine learning in Python", J Mach Learn Res, 12, pp.2825~2830 (2011)
- Chandrashekar G, "A survey on feature selection methods", Comput Electr Eng, 40(1), pp.16~28 (2014)
- Singh S, "Robust detection of Parkinson’s disease using harvested smartphone voice data : a telemedicine approach", Telemed J E Health, 26(3), pp.327~334 (2020)