Two-stage classification of respiratory sound patterns.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Medical Informatics

Background:

• Respiratory sound analysis is crucial for diagnosing pulmonary conditions.
• Traditional methods often overlook the cyclic nature of respiratory signals.
• Accurate classification of respiratory sounds remains a challenge.

Purpose of the Study:

• To develop a novel hierarchical decision fusion scheme for respiratory sound signal classification.
• To leverage the cyclic nature of respiration for improved diagnostic accuracy.
• To enhance the robustness of respiratory sound analysis.

Main Methods:

• Respiratory signals were partitioned into segments corresponding to six phases of the respiration cycle.
• Multilayer perceptron classifiers were employed to analyze parameterized segments from each phase.
• A nonlinear decision combination function integrated decisions from different phases.

Main Results:

• The hierarchical decision fusion scheme demonstrated effective classification of respiratory sound signals.
• Analyzing cyclic respiratory phases improved the overall classification performance.
• A new regularization scheme stabilized the training and consultation processes.

Conclusions:

• The proposed hierarchical decision fusion scheme offers a promising approach for respiratory sound classification.
• Incorporating the cyclic nature of respiration enhances diagnostic capabilities.
• The method provides a stable and accurate framework for analyzing respiratory signals.