EUVIMEDEUVIMED entdecken
Crossref

Harnessing Transformer-Based Language Models for Interstitial Lung Diseases Diagnosis From Time-Frequency Encodings of Lung Sounds

Ayushi Pal, Naseem Babu, Udit Satija, Jimson Mathew, Hugeng Hugeng, Choo W. R. Chiong

IEEE Sensors Letters · 2026-07 · 10 (7) · 6006404-6006404

DOI: 10.1109/lsens.2026.3702525 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Sensors Letters, 2026-07. Verfasst von Ayushi Pal und 5 weiteren Autor:innen. Die Arbeit nennt 20 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Early and accurate diagnosis of interstitial lung disease (ILD) remains a major clinical challenge due to its wide range of symptoms and overlapping sound patterns with other pulmonary disorders, such as fine crackles and squawks. This letter proposes a transformer-based language model (LM) framework for classifying respiratory sounds associated with ILD using three complementary time–frequency (TF) encodings: constant-Q transform, mel-spectrograms, and mel-frequency cepstral coefficients. Each TF representation is first derived from preprocessed lung sound recordings and then reshaped into token-like sequences suitable for processing by pretrained transformer backbones, specifically GPT-2, GPT-2-medium, Qwen, and Qwen-chat models. The proposed framework is evaluated on publicly available respiratory sound datasets, BRACETS and KAUH, and performance is reported as average accuracy, recall, precision, and F1-score to provide a comprehensive assessment. The proposed framework underscores the potential of transformer-based LM architectures operating on TF representations of respiratory sound (RS) as a promising direction for developing robust, noninvasive, and scalable tools for early ILD screening and for discovering reliable acoustic biomarkers in clinical and real-world applications.

Zitieren

APA

Pal, A., Babu, N., Satija, U., Mathew, J., Hugeng, H., & Chiong, C. W. R. (2026). Harnessing Transformer-Based Language Models for Interstitial Lung Diseases Diagnosis From Time-Frequency Encodings of Lung Sounds. IEEE Sensors Letters, 10(7), 6006404-6006404. https://doi.org/10.1109/lsens.2026.3702525

Vancouver

Pal A, Babu N, Satija U, Mathew J, Hugeng H, Chiong CWR. Harnessing Transformer-Based Language Models for Interstitial Lung Diseases Diagnosis From Time-Frequency Encodings of Lung Sounds. IEEE Sensors Letters. 2026;10(7):6006404-6006404. doi:10.1109/lsens.2026.3702525

BibTeX
@article{Pal2026,
  author = {Pal, Ayushi and Babu, Naseem and Satija, Udit and Mathew, Jimson and Hugeng, Hugeng and Chiong, Choo W. R.},
  title = {Harnessing Transformer-Based Language Models for Interstitial Lung Diseases Diagnosis From Time-Frequency Encodings of Lung Sounds},
  journal = {IEEE Sensors Letters},
  year = {2026},
  volume = {10},
  number = {7},
  pages = {6006404--6006404},
  doi = {10.1109/lsens.2026.3702525}
}
Kostenlos starten →