Customized Spectro-Temporal CNN Feature Extraction and ELM-Based Classifier for Accurate Respiratory Obstruction Detection

The accurate prediction based on lung auscultation of respiratory obstruction conditions (ROC), such as chronic obstructive pulmonary disease (COPD) and asthma, is a challenging task due to the availability of small datasets, ambient noise, variability between patients, high computational power, ove...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:IEEE access Ročník 13; s. 114145 - 114158
Hlavní autori: Muthulakshmi, M., Venkatesan, K., Bahiyah Rahayu, Syarifah, Nayana Sree, K. L.
Médium: Journal Article
Jazyk:English
Vydavateľské údaje: Piscataway IEEE 2025
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Predmet:
Accuracy
Artificial neural networks
Asthma
Audio data
Audio signals
Auscultation
Chronic obstructive pulmonary disease
CNN enhanced time-frequency features
Deep learning
ELM
Feature extraction
feature fusion
Health services
Low cost
Lungs
Machine learning
Mel frequency cepstral coefficient
Noise
Pneumonia
Real time
Representations
Respiratory diseases
respiratory obstruction condition
Spectrogram
Time-frequency analysis
ISSN:2169-3536, 2169-3536
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:The accurate prediction based on lung auscultation of respiratory obstruction conditions (ROC), such as chronic obstructive pulmonary disease (COPD) and asthma, is a challenging task due to the availability of small datasets, ambient noise, variability between patients, high computational power, overlapping auscultation characteristics and lack of universal clinical standards. Secondly, discrimination of obstructive respiratory diseases (COPD, asthma) and restrictive respiratory diseases (other) is critical, as they have different treatment and management strategies. Third, a clear distinction between COPD and asthma is of great concern, as treatment at the appropriate stage results in an open air passage in COPD and an improvement in shortness of breath in asthma. Although several techniques have been explored for diagnosing respiratory disease based on lung auscultation, there is still a need for an effective, low-cost, and faster solution suitable for real-time ROC detection. The time-frequency representations of audio signals are suitable to capture low-frequency information, as well as tonal and harmonic relationships. In contrast, deep learning architectures can learn complex, hierarchical, and high-level patterns from the spatio-temporal structures. In addition, extreme learning machines (ELM) can provide generalized performance with fewer training parameters. Hence, combining time-frequency representations, deep learning architecture, and ELM would result in the most reliable low-cost tool to predict ROC from lung auscultation. The fusion of deep features from different spatiotemporal structures outperforms individual features when fed into the ELM model, resulting in clear discrimination of obstructive and restrictive respiratory diseases. The proposed CNN-enhanced time-frequency features powered the ELM-based framework, yielding a test accuracy of 97.5% for the unseen test data considered. Thus, this study would be a useful aid for pulmonologists and would play a pivotal role in the accessibility to healthcare, early intervention, and long-term management of ROC disease.
Bibliografia:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:2169-3536
2169-3536
DOI:10.1109/ACCESS.2025.3581271