Cuffless Blood Pressure Estimation using PPG and ECG

This research marked a significant milestone in my transition from Nursing to Biostatistics, as it introduced me to the world of signal processing and machine learning for disease estimation. I am deeply grateful to Prof. ZHANG for his invaluable guidance and patience throughout this project. My work during this period encompassed three main components:

1. Replicating Research on Blood Pressure Estimation: In this part, I focused on replicating a paper that used machine learning models to estimate blood pressure (BP) based on photoplethysmogram (PPG) and electrocardiogram (ECG) signals. My tasks included:
   • Data Preprocessing: I worked with a dataset of 600 patients from the MIMIC-II database.
   • Feature Extraction: I extracted 13 essential features from the PPG and ECG signals, including pulse transit time (PTT), various amplitudes (T-wave, Q-wave, R-peak, S-peak, Low-peak), QRS intervals, R to low-peak amplitudes, R-peak intervals, PPG pulse amplitude, PPG pulse width, PPG high to low interval, and PPG slope change standard deviation.
   • Machine Learning Models: I applied several machine learning models in BP estimation, including Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Random Forest (RF), and Linear Models. Notably, the LSTM model outperformed the others.
2. Investigating the Relationship Between PPG, ECG, Reference BP, and Estimated BP:
   • I utilized LSTM and Random Forest (RF) models for estimating blood pressure waveforms.
   • I calculated the correlation coefficient (CC) for Reference BP & Estimated BP, PPG & Estimated BP, and ECG & Estimated BP.
3. Optimal Feature Set for Blood Pressure Estimation:
   • I Conducted a review of a published article: Beat-to-beat continuous blood pressure estimation with optimal feature set of PPG and ECG signals using deep recurrent neural networks.
   • I performed statistical analyses in this article.