Non-invasive continuous blood pressure estimation for athletes

athletes was developed using various machine learning models. Traditional BP measurement techniques are classified as invasive and non-invasive, both of which have certain limitations. In recent years, systems that estimate BP indirectly using biosignals such as electrocardiogram (ECG) and photoplethysmogram (PPG)—which can be easily measured from the body—have attracted significant attention as alternatives to overcome these disadvantages. In this study, several features were extracted: from the ECG signal—heart rate (HR), P-R interval, R-R interval, and the amplitudes of the P, R, and T waves; from the PPG signal—pulse period interval (PPI), systolic peak value (SPV), pulse wave end (PWE), and the ratio between the amplitude of the dicrotic notch (b) and the amplitude of the systolic peak (a).b/a ratio; and additionally, pulse transit time (PTT) was calculated using both signals. For the training and testing of the machine learning models, both publicly available datasets and original data collected via custom-designed hardware developed within the scope of this study were used. The designed system is capable of real-time signal acquisition, and the measurements were taken from subjects while running on a treadmill. Therefore, the system was specifically evaluated for athletic use, providing data that reflects BP variations during physical activity. Using the extracted features, several machine learning algorithms were tested for BP estimation. Among the tested models, the LASSO regression model achieved the best performance with a mean absolute error (MAE) of 2.3 ± 0.7 mmHg for diastolic BP and 5 ± 1.3 mmHg for systolic BP. The ECG and PPG signals used in the study were subjected to several preprocessing steps, including signal filtering, feature selection, and normalization, prior to training the models. In order to evaluate the generalization and accuracy of the models, different datasets were utilized during testing. Compared to other studies in the literature that estimate BP indirectly using ECG and PPG signals, this study developed a highly accurate algorithm using fewer features. In conclusion, the proposed system presents a promising approach for non-invasive and continuous blood pressure monitoring Keywords: Blood Pressure, ECG, PPG, Machine Learning