Accelerating Biomedical Signal Processing Using GPU: A Case Study of Snore Sound Feature Extraction.

Area of Science:

• Biomedical data analysis
• Machine learning in healthcare
• Personalized medicine

Background:

• Feature extraction is crucial for machine learning in biomedical data analysis.
• Handling 'Big Data' in healthcare presents significant computational challenges.
• Efficient feature extraction is essential for developing personalized healthcare solutions.

Purpose of the Study:

• To investigate the effectiveness of Graphics Processing Units (GPUs) for accelerating feature extraction from large biomedical datasets.
• To demonstrate a practical application of GPU-accelerated feature extraction in analyzing snore sound data.
• To facilitate the integration of extracted features into deep learning frameworks for personalized healthcare.

Main Methods:

• Utilized Python and Graphics Processing Units (GPUs) for feature extraction from a substantial corpus of snore sound data.
• Collected 17.20 GB of snore sound data from 20 subjects across multiple hospitals.
• Ensured extracted features were compatible with standard deep learning training frameworks.

Main Results:

• Achieved a significant speedup of up to seven times in the feature extraction phase compared to traditional Central Processing Unit (CPU) systems.
• Demonstrated the feasibility of GPU-based processing for large-scale biomedical data.
• Enabled direct import of extracted features into deep learning frameworks without format conversion.

Conclusions:

• GPU-based feature extraction offers a substantial performance improvement for analyzing large biomedical datasets.
• This approach can significantly reduce the time and computational resources required for personalized healthcare applications.
• Accelerated feature extraction paves the way for more efficient development and deployment of deep learning models in medicine.