Heartbeat Detection by Laser Doppler Vibrometry and Machine Learning

Area of Science:

• Cardiovascular physiology
• Biomedical engineering
• Machine learning applications

Background:

• Heartbeat detection is vital in clinical settings.
• Laser Doppler Vibrometer (LDV) offers non-contact measurement capabilities.
• Investigating machine learning for LDV-based heartbeat detection is warranted.

Purpose of the Study:

• To evaluate machine learning algorithms for heartbeat detection from carotid LDV signals.
• To compare the performance of Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), and K-Nearest Neighbor (KNN) classifiers.
• To assess the feasibility of real-world application for contactless cardiovascular signal analysis.

Main Methods:

• A dataset from 28 subjects was collected using LDV.
• Electrocardiography (ECG) served as the gold standard for labeling.
• Leave-one-subject-out cross-validation was employed to test classifier performance on signal windows.
• Classifiers were evaluated on their ability to distinguish between 'beat' and 'no-beat' windows.

Main Results:

• All tested machine learning classifiers achieved high f1-scores for the 'beat' class (0.93-0.96).
• Support Vector Machine (SVM) demonstrated the highest performance with an f1-score of 0.96.
• SVM achieved a median macro-f1 of 0.76 when tested on full-length LDV signals, simulating real-world conditions.
• No statistically significant differences were observed between the performances of the evaluated classifiers.

Conclusions:

• Machine learning models show significant potential for heartbeat detection using carotid LDV signals.
• This study represents a promising advancement in contactless cardiovascular signal analysis.
• The findings support the development of non-invasive methods for monitoring heart activity.