Figure 1.
Data cleaning and workflow process for developing machine learning models to estimate blood pressure using only ECGs. Abbreviations: valid label: ‘sinus rhythm’, ‘atrial fibrillation’, ‘bradycardia’, ‘tachycardia’, ‘too short’, ‘unclassified’, invalid label: ‘unreadable’ and missing values. Additionally, in the final stage up to 6 records per subject were selected, based on the median number of records per subject after excluding an outlier with 3500 records and subjects with only one record.