Cascade Convolutional Neural Network - Autoencoder Method for Parkinson’s Disease Detection Using Multilingual Voice Signal

Parkinson’s disease (PD) is a neurodegenerative disorder that hampers both motor and non-motor activities in human body. Speech signal has been used as a viable biomarker for early identification of PD. Voice changes in the temporal and spectral domains have been used to identify the condition effec...

Ausführliche Beschreibung

Gespeichert in:

Bibliographische Detailangaben
Veröffentlicht in:	SN computer science Jg. 6; H. 7; S. 802
Hauptverfasser:	Pandey, Pandit Vivek Kumar, Sahu, Sitanshu Sekhar
Format:	Journal Article
Sprache:	Englisch
Veröffentlicht:	Singapore Springer Nature Singapore 01.10.2025 Springer Nature B.V
Schlagworte:	Accuracy Artificial neural networks Biomarkers Computer Imaging Computer Science Computer Systems Organization and Communication Networks Data Structures and Information Theory Datasets Deep learning Dopamine English language Feature extraction Females Fourier transforms Human motion Information Systems and Communication Service Machine learning Males Multilingualism Neural networks Original Research Parkinson's disease Pattern Recognition and Graphics Software Engineering/Programming and Operating Systems Spanish language Speech Support vector machines Vision Voice Vowels Wavelet transforms United States > US CNN Bottleneck autoencoder SVM XGBoost Parkinson’s disease Spectrogram
ISSN:	2661-8907, 2662-995X, 2661-8907
Online-Zugang:	Volltext
Tags:	Tag hinzufügen Keine Tags, Fügen Sie den ersten Tag hinzu!

Beschreibung
Zusammenfassung:	Parkinson’s disease (PD) is a neurodegenerative disorder that hampers both motor and non-motor activities in human body. Speech signal has been used as a viable biomarker for early identification of PD. Voice changes in the temporal and spectral domains have been used to identify the condition effectively. This work introduces a unique cascaded CNN-bottleneck autoencoder-SVM model for efficient detection of PD. The proposed method extracts features from the voice spectrogram with a Convolutional Neural Network (CNN) to learn the latent feature representations. The large feature embedding of CNN is represented in a compact form using the autoencoder. The extracted features are fed to the classifier to precisely detect the PD and healthy. The efficiency of the proposed method is assessed using the sustained vowel of three different language datasets of Spanish, Italian and US English. In the language-specific performance, the proposed model achieved 80% highest accuracy in vowel /a/, /o/ and /u/ in dataset 1 (Spanish), in dataset 2 (Italian) achieved a maximum accuracy of 96% in vowel /u/ and in dataset 3 (US English), /a/ is achieved a maximum accuracy of 75%. When tested on cross dataset, it achieved highest accuracy of 68.33% for vowel /e/ (trained on Italian, tested on Spanish) and 63% for vowel /i/ (trained on Spanish, tested on Italian) and 56% (trained on US English, tested on Spanish) in vowel /a/. The experimental results highlight the potential of the method for effective detection of PD across multilingual datasets.
Bibliographie:	ObjectType-Article-1 SourceType-Scholarly Journals-1 ObjectType-Feature-2 content type line 14
ISSN:	2661-8907 2662-995X 2661-8907
DOI:	10.1007/s42979-025-04327-0