Preprocessing Methods for Ambulatory HRV Analysis Based on HRV Distribution, Variability and Characteristics (DVC).

Area of Science:

• Physiological signal processing
• Artificial Intelligence in Healthcare
• Wearable Technology Data Analysis

Background:

• Wearable devices and AI enable ambulatory physiological monitoring, including heart rate variability (HRV).
• Data quality issues like noise and missing values in ambulatory HRV recordings can compromise analysis, especially for medical diagnosis.
• Conventional interpolation methods for missing HRV data often fail to preserve crucial time-dependent signal characteristics.

Purpose of the Study:

• To introduce a novel HRV processing method incorporating filtering and iterative Gaussian-based data imputation.
• To evaluate the proposed method's effectiveness in preserving physiological and time-series characteristics of HRV signals.
• To compare the proposed method against traditional interpolation techniques for stress classification using a random forest classifier.

Main Methods:

• Development of an iterative data imputation technique using a Gaussian distribution, considering physiological aspects like HRV distribution, RR variability, and normal boundaries.
• Application of the proposed method and standard interpolation techniques (linear, pchip, spline) to reconstruct HRV signals from data with varying percentages of missing values.
• Classification of stress versus relaxation using a random forest algorithm on features extracted from the reconstructed HRV signals.

Main Results:

• The proposed iterative imputation method achieved a stable F1 score of 61% for stress classification, even with high percentages of missing data.
• Traditional interpolation methods yielded approximately 54% F1 score with low missing data percentages, dropping to around 44% with increased data loss.
• The proposed method demonstrated superior performance and robustness compared to other imputation techniques, particularly in scenarios with substantial data corruption.

Conclusions:

• The novel HRV processing method, utilizing iterative Gaussian imputation, offers significant advantages over conventional interpolation for signal reconstruction.
• This advanced imputation technique enhances the accuracy of stress classification from wearable-derived HRV data, especially when dealing with considerable data loss.
• The findings suggest the proposed method is a more reliable approach for analyzing ambulatory HRV data in real-world applications, including medical diagnosis.