Convolutional neural networks based efficient approach for classification of lung diseases

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Respiratory Medicine

Background:

• Lung diseases are a leading cause of mortality globally, necessitating advanced diagnostic tools.
• Automated lung sound analysis using artificial intelligence (AI) shows potential for assisting medical experts.
• Existing research utilizes stethoscope recordings for AI-based lung disease diagnosis.

Purpose of the Study:

• To classify lung sounds using two distinct deep learning approaches.
• To evaluate the performance of these AI models on the ICBHI 2017 lung sound database.
• To compare the proposed methods against existing benchmarks in lung sound classification.

Main Methods:

• Lung sound signals were transformed into spectrogram images using the Short-Time Fourier Transform (STFT).
• Approach 1: Employed a pre-trained deep Convolutional Neural Network (CNN) for feature extraction, followed by a Support Vector Machine (SVM) classifier.
• Approach 2: Utilized transfer learning by fine-tuning a pre-trained deep CNN model directly on spectrogram images for classification.

Main Results:

• The first approach (CNN + SVM) achieved an accuracy of 65.5% for lung sound classification.
• The second approach (fine-tuned CNN) yielded an accuracy of 63.09%.
• Both proposed methods demonstrated superior performance compared to existing results in the literature.

Conclusions:

• Deep learning models, particularly the CNN + SVM approach, show effectiveness in classifying lung sounds.
• The study highlights the potential of AI-driven analysis of respiratory sounds for improved diagnostic accuracy.
• Further research can build upon these findings to develop more robust AI tools for respiratory disease management.