Related Experiment Video
Updated: Oct 3, 2025
04:04
Asthma Detection Research Based on Voice Signal Processing and Machine Learning
Published on: July 22, 2025
483
Automated Lung Sound Classification Using a Hybrid CNN-LSTM Network and Focal Loss Function
Georgios Petmezas1, Grigorios-Aris Cheimariotis1, Leandros Stefanopoulos1
1Laboratory of Computing, Medical Informatics and Biomedical-Imaging Technologies, Medical School, Aristotle University of Thessaloniki, GR 54124 Thessaloniki, Greece.
Sensors (Basel, Switzerland)
|February 15, 2022
Summary
This study introduces a novel deep learning model for accurate respiratory disease diagnosis using lung sound classification. The hybrid model achieves state-of-the-art results, improving early detection and patient monitoring for respiratory conditions.
Area of Science:
- Medical Informatics
- Artificial Intelligence in Medicine
- Respiratory Medicine
Background:
- Respiratory diseases are a major global health concern, impacting quality of life.
- Accurate diagnosis and monitoring of respiratory conditions are crucial for effective patient management.
- Current methods like manual lung auscultation are subjective and require extensive expertise.
Purpose of the Study:
- To develop a robust deep learning model for automated lung sound classification.
- To address challenges in training data imbalance using a focal loss function.
- To improve the accuracy and efficiency of diagnosing respiratory diseases.
Main Methods:
- A hybrid neural network combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) was proposed.
- Features were extracted from Short-Time Fourier Transform (STFT) spectrograms using CNNs.
- The LSTM network processed temporal dependencies for classifying four lung sound types: normal, crackles, wheezes, and combined crackles/wheezes.
Main Results:
- The model achieved state-of-the-art performance on the ICBHI 2017 Respiratory Sound Database.
- Results varied across data splitting strategies, with notable accuracy and sensitivity reported.
- For the 60/40 split: sensitivity 47.37%, specificity 82.46%, score 64.92%, accuracy 73.69%.
Conclusions:
- The proposed hybrid deep learning model demonstrates significant potential for accurate lung sound classification.
- This approach offers a promising tool for objective and efficient diagnosis of respiratory diseases.
- Further validation across diverse datasets can enhance clinical applicability.
