AI-based approach for heart failure readmission prediction using SCG, ECG, and GSR signals

Area of Science:

• Cardiology
• Biomedical Engineering
• Data Science

Background:

• Heart failure (HF) is a global health crisis with increasing prevalence and significant economic impact.
• Predicting HF readmission is crucial for effective patient management and reducing healthcare burdens.

Purpose of the Study:

• To explore the potential of seismocardiogram (SCG) signals for noninvasively predicting HF patient readmission.
• To compare the efficacy of conventional machine learning (ML) and deep learning models in HF readmission prediction using SCG data.

Main Methods:

• Acquired SCG signals from 101 HF patients, including those with readmissions.
• Segmented SCG signals, extracted features, and developed ML models.
• Transformed SCG signals into images for deep learning model training.

Main Results:

• Machine learning models outperformed the deep learning model in classifying HF readmissions.
• K-nearest neighbor achieved the highest accuracy (89.4%), sensitivity (87.8%), and specificity (90.1%).
• Extracted SCG features were correlated with HF conditions, supporting their clinical relevance.

Conclusions:

• SCG signals represent a promising noninvasive tool for predicting HF patient readmission.
• ML-based analysis of SCG data offers a viable strategy for proactive HF management.