Voice analysis in asthmatic and healthy individuals: comparative assessment of asthma control levels using machine learning models


Alagöz N., Uysal M. A., Gezer M., Ortak C., NİKSARLIOĞLU E. Y.

BMC Pulmonary Medicine, cilt.26, sa.1, 2026 (SCI-Expanded, Scopus)

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 26 Sayı: 1
  • Basım Tarihi: 2026
  • Doi Numarası: 10.1186/s12890-026-04654-9
  • Dergi Adı: BMC Pulmonary Medicine
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, CINAHL, EMBASE, MEDLINE, Directory of Open Access Journals, Biomedical Reference Collection: Corporate Edition (EBSCO), Health Research Premium Collection (ProQuest)
  • Anahtar Kelimeler: Asthma diagnosis, Asthma control, Artificial intelligence, Machine learning
  • Sağlık Bilimleri Üniversitesi Adresli: Evet

Özet

Introduction: Machine learning (ML) has significant potential for diagnosing and monitoring asthma. Previous studies have primarily focused on wheezing, coughing, breathing sounds, and phonetic features, while the potential of specific spoken words remains less explored. This study aimed to use ML models to distinguish individuals with asthma from healthy controls and classify asthma control according to the Global Initiative for Asthma (GINA) criteria using recorded vowels and specific words. Methods: This single-center, cross-sectional study included 344 participants: 284 patients with asthma and 60 healthy volunteers. Participants were asked to repeat six different Turkish words 10 times each and sustain the vowel “a” for 10 s. Voice recordings were analyzed using various ML models, including ensemble methods, Support Vector Machines (SVM), and multilayer perceptrons (MLP). The model performance was evaluated based on metrics, including accuracy, sensitivity, and F1 score. Results: When distinguishing asthmatic patients from healthy individuals, the highest-performing model (Voting Ensemble using the word ‘gelecek) achieved an AUC of 0.769, accuracy of 82.0%, and sensitivity of 93.3%. For the classification of asthma control status, the models showed limited discriminatory power, with the highest AUC of 0.592, reflecting significant class imbalance between controlled and uncontrolled patients. Conclusions: Speech-based machine learning models have shown potential as a noninvasive approach for distinguishing individuals with asthma from healthy controls. Therefore, voice-based assessment may serve as a complementary tool to standard clinical and functional evaluations. Larger, more diverse populations and external validation are required before clinical implementation.