Wearable Devices, Smartphones, and Interpretable Artificial Intelligence in Combating COVID-19

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Epidemiology

Background:

• Physiological measures like heart rate variability (HRV) and beats per minute (BPM) are key indicators of respiratory infections.
• Biometric wearables and smartphones can easily collect HRV and BPM data.
• Abnormal changes in these metrics may signal early stages of infections like COVID-19.

Purpose of the Study:

• To investigate the utility of heart measurements from wearables and smartphones for early detection of COVID-19 related inflammation.
• To develop and evaluate an AI framework for identifying potential COVID-19 infections based on physiological and textual data.

Main Methods:

• An AI framework combining an interpretable prediction model for HRV status and a recurrent neural network (RNN) for analyzing daily user logs.
• Utilizing a public dataset of 186 patients with over 3200 HRV readings and textual logs.
• Employing Local Interpretable Model-agnostic Explanations (LIME) for model interpretability.

Main Results:

• The AI approach achieved an accuracy of 83.34 ± 1.68% in predicting infection.
• The model demonstrated strong performance with a precision of 0.91, recall of 0.88, and F1-Score of 0.89.
• The system successfully identified potential infections up to two days prior to symptom onset.

Conclusions:

• Heart measurements from wearables and smartphones, analyzed by AI, show significant potential for early COVID-19 detection.
• The developed AI framework effectively integrates physiological and contextual data for infection prediction.
• This approach supports the use of consumer technology in public health surveillance and combating infectious diseases.