Machine Learning Based Classification of Myocardial Infarction Conditions Using Smartphone-Derived Seismo- and Gyrocardiography
Executive Summary
This study explores the feasibility of using smartphone-derived seismocardiography (SCG) and gyrocardiography (GCG) signals to classify pre- and post-operation cardiac conditions in ST-elevation myocardial infarction (STEMI) patients. Data from 20 patients were processed using a two-stage artifact removal method and 25 features were extracted for classification using Random Forest and Support Vector Machine algorithms, achieving accuracy rates of 74% and 78%, respectively. The findings suggest potential for low-cost, accessible remote cardiac monitoring systems for post-PCI patients.
Answer Machine Insights
Q: What was the primary method used for artifact removal?
A two-stage non-parametric artifact removal method using segmentation and Shannon entropy estimation.
A two-stage non-parametric artefact removal was adopted in this study mainly because the signals contain artefacts with different probability distributions.
Q: What features were extracted from the SCG and GCG signals?
25 features including mean, standard deviation, skewness, kurtosis, dominant frequency, and spectral centroid.
After transforming the artefact-free signals to zero-mean and unit variance, a combination of statistical and signal processing features were extracted from all the signals.
Key Results
Random Forest classifier achieved 74% accuracy, with 78% specificity and 70% sensitivity.
Support Vector Machine classifier achieved 78% accuracy, with 75% specificity and 80% sensitivity.
Clinical Snapshot
Evidence Rating
Relevance
high Priority