Smartphone Based Human Breath Analysis from Respiratory Sounds.

Area of Science:

• Biomedical Engineering
• Pulmonary Medicine
• Signal Processing

Background:

• Human breath analysis is crucial for diagnosing and managing pulmonary diseases.
• Distinguishing between normal and abnormal lung sounds is a critical diagnostic challenge.
• Respiratory diseases often manifest as irregular patterns in respiratory cycles.

Purpose of the Study:

• To develop and validate a scheme for breath analysis to detect irregular patterns indicative of respiratory diseases.
• To assess the feasibility of using smartphone-captured breath sounds for diagnostic purposes.
• To evaluate the classification performance of the proposed method for asthmatic and complete respiratory sound cycles.

Main Methods:

• Breath segments were de-noised using a wavelet de-noising method.
• Intrinsic mode functions were extracted using Complete Ensemble Empirical Mode Decomposition (CEEMD).
• Instantaneous frequency (IF) and envelope were extracted for feature extraction, followed by Bag-of-word grouping and Support Vector Machine (SVM) classification.

Main Results:

• The study analyzed 255 breath cycles captured via smartphone.
• Classification accuracy achieved was 75.21%±2% for asthmatic inspiratory cycles.
• Accuracy for complete Respiratory Sounds (RS) cycles was 75.5%±3%, with diagnostic odds ratios of 20.61% and 13.57% respectively.

Conclusions:

• The proposed breath analysis scheme demonstrates potential for detecting respiratory abnormalities using smartphone data.
• The combination of CEEMD, IF/envelope extraction, and SVM offers a robust approach for respiratory sound classification.
• Further research can refine this method for improved diagnostic accuracy in pulmonary disease management.