Parkinson's Disease Detection from Speech Signals: Comparison of Three Different Approaches with Cross-Corpus Evaluation

Öz

This study presents a cross-corpus evaluation of three different approaches for the detection of Parkinson's disease (PD) from speech signals. Three publicly available datasets (PC-GITA, MDVR-KCL, and mPower), each with different recording environments, languages, and speech tasks, were used, and model generalizability was assessed using a Leave-One-Dataset-Out (LODO) protocol. Three approaches were compared: (i) a Mel-spectrogram-based Hybrid CNN+ViT (EfficientNetB0 + Vision Transformer) deep learning model, (ii) a Wav2Vec 2.0-based transfer learning approach, and (iii) classical ML models (SVM, KNN, MLP, RF) trained on ViT-PCA combined with acoustic features (MFCC, delta, and delta-delta). Results were reported at both segment and subject levels, and the effect of segment duration (10, 15, 20, and 25 seconds) on performance was systematically investigated. In the S1 scenario (Test: mPower), the MLP model achieved the highest accuracy (88.91%) with 20-second segments, whereas the RF model obtained the highest AUC (0.961). The Hybrid CNN+ViT model attained an AUC of 0.935 and 85.81% accuracy with 15-second segments. In the S3 scenario (Test: PC-GITA), the Wav2Vec 2.0-based Deep Neural Network (DNN) model achieved an AUC of 0.726, outperforming all spectrogram-based approaches. The findings suggest that speech analysis may be considered for use in Parkinson’s disease screening applications; however, generalization across different speech tasks remains a significant open challenge.

Anahtar Kelimeler

• Parkinson's disease
• Cross-corpus evaluation
• Vision Transformer
• Mel-spectrogram
• Wav2Vec 2.0
• Transfer learning
• Deep learning

Destekleyen Kurum

This work was not supported by any institution or funding agency. No individuals or organizations contributed to this study.

Etik Beyan

I declare that all processes of this study comply with research and publication ethics, and that I have adhered to ethical guidelines and principles of scientific citation.

Teşekkür

This work was not supported by any institution or funding agency. No individuals or organizations contributed to this study

Konuşma Sinyallerinden Parkinson Hastalığı Tespiti: Çapraz Veri Kümesi Değerlendirmesi ile Üç Farklı Yaklaşımın Karşılaştırılması

Öz

Bu çalışma, konuşma sinyallerinden Parkinson hastalığının (PH) tespiti için üç farklı yaklaşımın çapraz veri kümesi (cross-corpus) değerlendirmesini sunmaktadır. Farklı kayıt ortamları, diller ve konuşma görevleri içeren üç açık erişimli veri seti (PC-GITA, MDVR-KCL ve mPower) kullanılmış ve modellerin genellenebilirliği Leave-One-Dataset-Out (LODO) protokolü ile değerlendirilmiştir. Çalışmada üç yaklaşım karşılaştırılmıştır: (i) Mel-spektrogram tabanlı Hibrit CNN+ViT (EfficientNetB0 + Vision Transformer) derin öğrenme modeli, (ii) Wav2Vec 2.0 tabanlı transfer öğrenme yaklaşımı ve (iii) ViT-PCA ile akustik özelliklerin (MFCC, delta ve delta-delta) birleşiminden elde edilen öznitelikler üzerinde eğitilen klasik makine öğrenmesi modelleri (SVM, KNN, MLP, RF). Sonuçlar hem segment düzeyinde hem de birey (subject) düzeyinde raporlanmış, ayrıca segment süresinin (10, 15, 20 ve 25 saniye) performans üzerindeki etkisi sistematik olarak incelenmiştir. S1 senaryosunda (Test: mPower), MLP modeli 20 saniyelik segmentlerde en yüksek doğruluk oranını (%88,91) elde ederken, RF modeli en yüksek AUC değerine (0,961) ulaşmıştır. Hibrit CNN+ViT modeli ise 15 saniyelik segmentlerde 0,935 AUC ve %85,81 doğruluk elde etmiştir. S3 senaryosunda (Test: PC-GITA), Wav2Vec 2.0 tabanlı Derin Sinir Ağı (DNN) modeli 0,726 AUC değeri ile tüm spektrogram tabanlı yaklaşımlardan daha yüksek performans göstermiştir. Elde edilen bulgular, konuşma analizinin Parkinson hastalığının ön tarama uygulamalarında değerlendirilebileceğini göstermektedir; ancak farklı konuşma görevleri arasında genellenebilirlik hâlen önemli bir araştırma problemi olarak varlığını sürdürmektedir.

Anahtar Kelimeler

• Parkinson hastalığı
• Çapraz veri kümesi değerlendirmesi
• Vision Transformer
• Mel-spektrogram
• Wav2Vec 2.0
• Transfer öğrenme
• Derin öğrenme

Destekleyen Kurum

Bu çalışma herhangi bir kurum veya finansman kuruluşu tarafından desteklenmemiştir. Bu çalışmaya herhangi bir kişi veya kuruluş katkıda bulunmamıştır.

Etik Beyan

Bu çalışmanın tüm süreçlerinin araştırma ve yayın etiğine uygun olduğunu, etik kurallara ve bilimsel atıf ilkelerine bağlı kaldığımı beyan ederim.

Teşekkür

Bu çalışma herhangi bir kurum veya finansman kuruluşu tarafından desteklenmemiştir. Bu çalışmaya herhangi bir kişi veya kuruluş katkıda bulunmamıştır.

Kaynakça

1. T. Pringsheim, N. Jette, A. Frolkis, and T. D. L. Steeves, "The Prevalence of Parkinson's Disease: A Systematic Review and Meta-analysis," Movement Disorders, vol. 29, no. 13, pp. 1583–1590, 2014.
2. H. Gümüş, Z. Akpınar, and O. Demir, "Assessment of Early Stage Non-Motor Symptoms in Parkinson’s Disease," Turkish Journal of Neurology, vol. 19, no. 2, pp. 97–103, 2013.
3. L. Zahid et al., "A Spectrogram-Based Deep Feature Assisted Computer-Aided Diagnostic System for Parkinson’s Disease," IEEE Access, vol. 8, pp. 35482–35495, 2020, doi: 10.1109/ACCESS.2020.2974008.
4. A. Tsanas, M. A. Little, P. E. McSharry, and L. O. Ramig, “Accurate telemonitoring of Parkinson’s disease progression by noninvasive speech tests,” IEEE Transactions on Biomedical Engineering, vol. 57, no. 4, pp. 884–893, Apr. 2010.
5. S. Zhang, D. Zhang, and Y. Wang, “Deep learning-based approach for Parkinson’s disease detection using voice signals,” IEEE Access, vol. 10, pp. 11845–11855, 2022.
6. E. J. Ibarra, J. D. Arias-Londoño, M. Zañartu, and J. I. Godino-Llorente, "Towards a Corpus (and Language)- Independent Screening of Parkinson's Disease from Voice and Speech through Domain Adaptation," Bioengineering, vol. 10, no. 11, p. 1316, Nov. 2023, doi: 10.3390/bioengineering10111316.
7. M. Meral, F. Ozbilgin, and F. Durmus, “Fine-Tuned Machine Learning Classifiers for Diagnosing Parkinson’s Disease Using Vocal Characteristics: A Comparative Analysis,” Diagnostics, vol. 15, no. 5, p. 645, Mar. 2025, doi: 10.3390/diagnostics15050645.
8. B. Çelik and A. Akbal, "Distinguishing Parkinson's Patients Using Voice-Based Feature Extraction and Classification," arXiv preprint arXiv:2501.14390, Jan. 2025.

9. Halil GÖKÇİN, “Comparison of Machine Learning Algorithms Using Sound Signals in Detection of Parkinson’s Disease,” Harran Üniversitesi Mühendislik Derg., vol. 8, no. 2, pp. 130–140, 2023, doi: 10.46578/humder.1217255.
10. S.-M. Jeong, S. Kim, E. C. Lee, and H. J. Kim, "Exploring Spectrogram-Based Audio Classification for Parkinson's Disease: A Study on Speech Classification and Qualitative Reliability Verification," Sensors, vol. 24, no. 14, p. 4625, Jul. 2024, doi: 10.3390/s24144625
11. N. Madusanka and B. il Lee, “Vocal Biomarkers for Parkinson’s Disease Classification Using Audio Spectrogram Transformers,” J. Voice, 2024, doi: 10.1016/j.jvoice.2024.11.008.
12. M. B. Er, E. Isik, and I. Isik, “Parkinson’s detection based on combined CNN and LSTM using enhanced speech signals with Variational mode decomposition,” Biomed. Signal Process. Control, vol. 70, 2021, doi: 10.1016/j.bspc.2021.103006.
13. M. La Quatra, J. R. Orozco-Arroyave, and M. S. Siniscalchi, "Bilingual dual-head deep model for Parkinson's disease detection from speech," in Proc. ICASSP 2025 – IEEE Int. Conf. Acoust., Speech Signal Process., Hyderabad, India, 2025, pp. 1–5, doi: 10.1109/ICASSP49660.2025.10889445.
14. H. Sedigh Malekroodi, B. Il Lee, and M. Yi, “Voice-Based Detection of Parkinson’s Disease Using Machine and Deep Learning Approaches: A Systematic Review,” Bioengineering, vol. 12, no. 11, pp. 1–29, 2025, doi: 10.3390/bioengineering12111279.
15. B. Perrone, F. Amato, and G. Olmo, "Voice classification in Parkinson's disease: A deep learning approach using transformers and error rate metrics," Biomedical Signal Processing and Control, vol. 113, p. 108954, 2026, doi: 10.1016/j.bspc.2025.108954.
16. I. Tougui, M. Zakroum, O. Karrakchou, and M. Ghogho, "Transformer-based transfer learning on self-reported voice recordings for Parkinson's disease diagnosis," Scientific Reports, vol. 14, no. 1, p. 30131, Dec. 2024, doi: 10.1038/s41598-024-81824-x.
17. S.-M. Jeong, Y. D. Song, C. L. Seok, J.-Y. Lee, E. C. Lee, and H.-J. Kim, "Machine learning-based classification of Parkinson's disease using acoustic features: Insights from multilingual speech tasks," Computers in Biology and Medicine, vol. 182, p. 109078, 2024, doi: 10.1016/j.compbiomed.2024.109078.
18. M. S. Ahmed, P. Tithi, and M. Rahman, "Parkinson's disease detection using deep learning method & explainable AI," in Proc. 10th IEEE WIECON-ECE 2024, Dhaka, Bangladesh, 2024, pp. 1–6, doi: 10.1109/WIECON- ECE62528.2024.10915175.
19. M. Siniukov, E. Xing, S. Attaripour Isfahani, and M. Soleymani, "Towards a Generalizable Speech Marker for Parkinson's Disease Diagnosis," arXiv preprint arXiv:2501.03581, Jan. 2025.
20. J. R. Orozco-Arroyave, J. D. Arias-Londoño, J. F. Vargas-Bonilla, M. C. Gonzalez-Rativa, and E. Nöth, "New Spanish speech corpus database for the analysis of people suffering from Parkinson's disease," in Proc. 9th Int. Conf. Language Resources and Evaluation (LREC), Reykjavik, Iceland, 2014, pp. 342–347.
21. MDVR-KCL Dataset. Kaggle. [Online]. Available: https://www.kaggle.com/datasets/nutansingh/mdvr-kcl-dataset. Accessed: Feb. 21, 2026.
22. B. M. Bot, C. Suver, E. C. Neto, M. Kellen, A. Klein, C. Bare, M. Doerr, A. Pratap, J. Wilbanks, E. R. Dorsey, S. H. Friend, and A. D. Trister, "The mPower study, Parkinson disease mobile data collected using ResearchKit," Scientific Data, vol. 3, p. 160011, Mar. 2016, doi: 10.1038/sdata.2016.11.