Processed MIMIC-III Waveform Dataset for ABP-ECG Modeling

This dataset is a preprocessed derivative of the MIMIC-III Waveform Database (PhysioNet), containing paired Arterial Blood Pressure (ABP) and Electrocardiogram (ECG) segments for physiological signal modeling. ABP and ECG signals were extracted from 895 ICU patients and processed through multi-stage quality control, including missing value removal, ABP range validation, signal standardization, and peak-based segment filtering. Signals were resampled from 125 Hz to 100 Hz and divided into 10-second non-overlapping windows, yielding 3,063,004 segments. For each signal type, 470,000 segments were selected across three splits: - Training set : 400,000 segments - Validation set: 50,000 segments - Test set : 20,000 segments This dataset is released under the ODC Open Database License (ODbL) v1.0 in accordance with the Share-Alike terms of the original database. Original source:Moody et al. (2020). MIMIC-III Waveform Database (v1.0). PhysioNet. https://doi.org/10.13026/c2607m