Feature extraction techniques for low-power ambulatory wheeze detection wearables

Area of Science:

• Medical Technology
• Pulmonology
• Signal Processing

Background:

• Wheezing in breathing sounds is a key indicator of various respiratory and pulmonary diseases.
• Accurate and efficient wheeze detection is crucial for diagnosing and monitoring lung conditions.

Purpose of the Study:

• To develop a novel, low-complexity algorithm for automatic wheeze detection.
• To propose hardware-friendly variants suitable for wearable devices.
• To evaluate the computational efficiency and power consumption of the proposed method.

Main Methods:

• A novel wheeze detection algorithm based on frequency contour tracking was developed.
• Two hardware-friendly variants of the algorithm were proposed.
• The algorithm's performance was evaluated using classification accuracy, computational complexity, and power consumption metrics.

Main Results:

• The proposed feature extraction algorithm achieved very high classification accuracy (>99%).
• The method demonstrated considerably low computational complexity (3×-6×) compared to previous approaches.
• Power consumption was significantly reduced (70×-100×) compared to 'record and transmit' strategies in wearable devices.

Conclusions:

• The developed low-complexity wheeze detection method offers high accuracy and efficiency.
• The hardware-friendly variants are suitable for implementation in resource-constrained wearable devices.
• This approach presents a significant advancement for remote respiratory monitoring and diagnostics.