A Pulse Signal Preprocessing Method Based on the Chauvenet Criterion.

Area of Science:

• Biomedical Signal Processing
• Cardiovascular and Respiratory Monitoring
• Machine Learning in Healthcare

Background:

• Pulse signals are crucial for assessing cardiovascular, respiratory, and circulatory health.
• Spike and poor-sensor-contact noise significantly degrade pulse signal accuracy.
• Existing noise reduction methods are often complex and difficult to implement.

Purpose of the Study:

• To propose a novel, efficient pulse signal preprocessing method based on the Chauvenet criterion.
• To effectively discriminate and remove abnormal signals caused by spike and poor-sensor-contact noise.
• To enhance the accuracy of subsequent detection models, such as those for sleep apnea.

Main Methods:

• Development of a pulse signal preprocessing technique utilizing the Chauvenet criterion.
• Implementation of adaptive thresholds for discriminating abnormal signals.
• Application and evaluation on 81 hours of pulse signals from the MIT-BIH Polysomnographic Database.

Main Results:

• The proposed method achieved 99.63% accuracy in discriminating 9,684 out of 9,720 noisy signal segments.
• Quantitative evaluation showed improved signal quality with a higher Jaccard Similarity Coefficient (JSC) post-processing.
• Back-propagation sleep apnea detection models using the preprocessed signals demonstrated higher recognition and prediction rates.
• The Chauvenet-based method exhibited significantly shorter execution times compared to the Romanovsky-based method, especially with larger datasets.

Conclusions:

• The proposed Chauvenet criterion-based method offers an effective and computationally efficient solution for pulse signal noise reduction.
• This preprocessing technique demonstrably improves signal quality and enhances the performance of sleep apnea detection models.
• The method's simplicity and speed make it a valuable tool for real-time biomedical signal analysis.