Evaluation of Deep Learning Methods for Pulmonary Disease Classification.

Area of Science:

• Medical Diagnostics
• Artificial Intelligence in Healthcare
• Signal Processing

Background:

• Rising prevalence of lung conditions due to pollution and infections necessitates improved diagnostic tools.
• Current diagnostic methods require advancements for greater accuracy and efficiency.

Purpose of the Study:

• To develop and evaluate a multi-feature deep learning model for enhanced pulmonary disease classification from auscultation recordings.
• To improve disease detection accuracy by integrating diverse audio features and advanced neural network architectures.

Main Methods:

• Extraction of audio features including spectrograms, chromograms, and Mel Frequency Cepstral Coefficients (MFCC).
• Application of filter-based audio enhancement techniques to mitigate background noise.
• Utilizing Convolutional Neural Networks (CNNs) for feature extraction and dense neural networks for classification.

Main Results:

• Deep learning models (CNN, RNN) achieved 70-85% accuracy, outperforming traditional machine learning.
• A combined CNN, RNN, and Long Short-Term Memory model reached 88% accuracy.
• The proposed multi-feature deep learning model integrating MFCC, Chroma STFT, and spectrograms achieved a highest accuracy of 92%.

Conclusions:

• The study successfully developed a high-accuracy model for pulmonary disease classification, achieving 92% accuracy.
• The methodology addresses common challenges such as background noise and imbalanced datasets.
• Findings demonstrate the potential of multi-feature deep learning for improved clinical applications in diagnosing lung diseases.