Robust_BP_BenchMark

OverviewRobust-BP-Bench is a highly curated, balanced, and high-quality subset derived from the MIMIC-II waveform database, specifically designed for continuous non-invasive blood pressure (cNIBP) estimation and physiological signal processing tasks.

MotivationOne of the major bottlenecks in developing Machine Learning / Deep Learning models for BP estimation is the extreme class imbalance in the raw MIMIC-II database (dominated by normal blood pressure ranges) and the presence of severe sensor artifacts. To address this, we developed a rigorous 4-bin stratified sampling protocol to ensure robust representation across all blood pressure conditions, particularly high-fluctuation and hypertensive events.

Dataset Characteristics- Signals Included: Photoplethysmogram (PPG), Electrocardiogram (ECG), and Arterial Blood Pressure (ABP).- Sampling Rate: 200 Hz.- Balanced Distribution: The dataset is strictly stratified into 4 bins based on Systolic Blood Pressure (SBP) to prevent model bias:  * Hypotension / Ideal (< 110 mmHg)  * Normal (110 - 130 mmHg)  * Pre-hypertension (130 - 150 mmHg)  * Hypertension (> 150 mmHg)- Strict Quality Control: All segments have passed rigorous filtering, including variance checks and linearity verifications, ensuring high Signal-to-Noise Ratio (SNR) for immediate model training.

Code & ReproducibilityThe open-source MATLAB scripts used for data selection, signal preprocessing, and baseline feature extraction are fully available on our GitHub repository:[https://github.com/phish-tech/Robust_Blood_Pressure_Benchmark]

(Note: Please replace the GitHub link above with your actual repository URL once public).