A Deep Ensemble Neural Network with Attention Mechanisms for Lung Abnormality Classification Using Audio Inputs.

Area of Science:

• Medical acoustics
• Artificial Intelligence in Medicine
• Respiratory Medicine

Background:

• Diagnosing lung abnormalities from audio is challenging due to unstructured respiratory signals.
• Existing methods often struggle with the complexity of respiratory sound classification.

Purpose of the Study:

• To develop an advanced ensemble model for diagnosing lung abnormalities and COVID-19 using diverse audio inputs.
• To improve the accuracy of medical audio classification for respiratory diseases.

Main Methods:

• An ensemble model integrating four deep neural networks: attention-based Convolutional Recurrent Neural Network (A-CRNN), attention-based bidirectional Long Short-Term Memory (A-BiLSTM), attention-based bidirectional Gated Recurrent Unit (A-BiGRU), and Convolutional Neural Network (CNN).
• Particle Swarm Optimization (PSO) for optimizing network training parameters.
• Ensemble mechanism combining probability predictions from base networks via averaging.

Main Results:

• The ensemble model and base networks achieved high performance, with ICBHI scores ranging from 0.920 to 0.9766 across various datasets.
• Empirical results demonstrate the model's capability to effectively distinguish COVID-19 positive cases from other common respiratory diseases using audio recordings.

Conclusions:

• The proposed ensemble deep learning model shows significant promise for accurate lung abnormality and COVID-19 diagnosis using respiratory, speech, and cough audio.
• Audio analysis offers a viable, non-invasive method for differentiating COVID-19 from other respiratory illnesses.