EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Mel-TansWeD-CNN: Respiratory Diseases Detection Framework Through Acoustic Features and Fusion of CNN with Transformer

Auliya Ur Rahman, Asima Ismail

International Journal of Innovations in Science and Technology · 15.07.2026 · 1604

DOI: 10.33411/ijist/1957 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Innovations in Science and Technology, online erschienen am 15.07.2026. Verfasst von Auliya Ur Rahman und 1 weiteren Autor:innen. Die Arbeit nennt 30 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Respiratory diseases represent a significant global health challenge, which requires a reliable framework for early diagnosis detection framework to assist clinicians in making their clinical decisions. Elderly people and people living in industrial areas are commonly most affected due to pollution. In this study, we designed Mel-TransWed-CNN, a hybrid deep learning (DL) framework that fuses Mel-frequency cepstral coefficients (MFCC) with Transformer and Convolutional Neural Network (CNN) to capture local spectral and global temporal patterns, respectively. We used the standard ICBHI 2017 respiratory sound dataset for experiments. The proposed approach consists of three main stages. First, the raw audio recordings are resampled and normalized to ensure consistency and improve signal quality. Next, MFCC with 20 coefficients are extracted to capture discriminative acoustic characteristics of respiratory sounds. Afterward, the extracted features are fed into the proposed fusion architecture, which combines a CNN with a transformer, where an initial CNN layer learns local spectral patterns, followed by a transformer encoder that models long-range temporal dependencies. A subsequent CNN stage further refines representations before pooling and Softmax-based multiclass classification. Furthermore, the proposed Mel-TransWeD-CNN achieved a mean accuracy of 95% with a standard deviation of 0.64% and a 95% confidence interval of 92.91%–95.43%, while outperforming the baseline classifiers, including Naive Bayes (88.0%), Gaussian Process Classifier (87.5%), Nearest Centroid (77.9%), and AdaBoost (71.7%). The proposed method contributes a practical and reliable automated framework for respiratory disease detection, with potential for early diagnosis and remote patient monitoring.

Zitieren

APA

Rahman, A. U., & Ismail, A. (2026). Mel-TansWeD-CNN: Respiratory Diseases Detection Framework Through Acoustic Features and Fusion of CNN with Transformer. International Journal of Innovations in Science and Technology, 1604. https://doi.org/10.33411/ijist/1957

Vancouver

Rahman AU, Ismail A. Mel-TansWeD-CNN: Respiratory Diseases Detection Framework Through Acoustic Features and Fusion of CNN with Transformer. International Journal of Innovations in Science and Technology. 2026:1604. doi:10.33411/ijist/1957

BibTeX
@article{Rahman2026,
  author = {Rahman, Auliya Ur and Ismail, Asima},
  title = {Mel-TansWeD-CNN: Respiratory Diseases Detection Framework Through Acoustic Features and Fusion of CNN with Transformer},
  journal = {International Journal of Innovations in Science and Technology},
  year = {2026},
  pages = {1604},
  doi = {10.33411/ijist/1957}
}
Kostenlos starten →