Automatic Multi-Level In-Exhale Segmentation and Enhanced Generalized S-Transform for wheezing detection.

Area of Science:

• Medical Informatics
• Signal Processing
• Machine Learning

Background:

• Wheezing is a common respiratory symptom indicative of asthma and COPD.
• Accurate wheezing detection aids in disease diagnosis, monitoring, and treatment.
• Automated methods offer objective assessment of lung sounds compared to traditional techniques.

Purpose of the Study:

• To develop an innovative machine learning-based approach for automated wheezing detection.
• To enhance the accuracy of wheezing classification by automatically segmenting respiratory sound phases and extracting relevant features.

Main Methods:

• Proposed Adaptive Multi-Level In-Exhale Segmentation (AMIE_SEG) for precise respiratory phase segmentation.
• Introduced Enhanced Generalized S-Transform (EGST) for effective wheezing feature extraction.
• Utilized machine learning classifiers (SVM, ELM, KNN) for wheezing detection.

Main Results:

• The proposed method achieved high performance across multiple classifiers on public datasets.
• At the segment level, KNN classifier demonstrated excellent results with 98.62% accuracy, 95.9% sensitivity, and 99.3% specificity.
• At the record level, classifiers achieved up to 99.52% accuracy, 100% sensitivity, and 99.27% specificity.

Conclusions:

• The developed machine learning approach shows significant promise for accurate wheezing detection.
• The methods are well-suited for long-term respiratory monitoring and telemedicine applications.