This project aimed to develop a machine-learning model capable of predicting blood pressure levels based on features extracted from electrocardiogram (ECG) signals. The goal was to explore the potential of non-invasive ECG monitoring as a means of predicting blood pressure without the need for traditional cuff-based measurements.
Data Acquisition and Preprocessing
Data Source: Kaggle
The main goal of this data set is to provide clean and valid signals for designing cuff-less blood pressure estimation algorithms. The raw electrocardiogram (ECG), photoplethysmograph (PPG), and arterial blood pressure (ABP) signals are originally collected from physionet.org and then some preprocessing and validation are performed on them.
This database consists of a cell array of matrices, each cell is one record part. In each matrix, each row corresponds to one signal channel:
1: PPG signal, FS=125Hz; photoplethysmograph from fingertip
2: ABP signal, FS=125Hz; invasive arterial blood pressure (mmHg)
3: ECG signal, FS=125Hz; electrocardiogram from channel II
The dataset is split into multiple parts to make it easier to load on machines with low memory. Each cell is a record.
Preprocessing:
Reshaped the ECG, PPG, and BP signal data into column vectors.
Exploratory Data Analysis (EDA)
Plotted sample PPG, ECG, and BP signals. Computed and plotted the cross-correlation between BP and ECG. Visualized the training error for each model tested.
Methodology and Modeling
- Went through related research papers and tested different models to predict BP from ECG.
- Tested several neural networks and machine learning models by changing configurations and parameters.
- Visualized the training error for each model tested.
Results and Evaluation
After evaluating different models I got better results while using both ECG and PPG to make the prediction.
Conclusion
The blood pressure prediction project leveraged advanced machine learning techniques and signal processing methods to develop a novel approach for non-invasive blood pressure monitoring using ECG signals. By harnessing the predictive power of ECG-derived features, the project has the potential to revolutionize cardiovascular health monitoring and contribute to proactive healthcare management strategies.
