Signal processing methods applied to ballistocardiogram signals for heart rate feature extraction

Nowadays, cardiac diseases are the major cause of death in the world, so to help in their diagnosis and monitoring non-obtrusive techniques are being explored and studied to have a more comfortable experience for the patients. The signal leading these studies is ballistocardiogram (BCG), which relies on the pressure exerted by the blood when travelling through the vascular tree. The present thesis investigates the extraction of heart rate features from BCG signals in optimal working conditions. Conducted at Holst Centre, this research capitalizes on a sensored mat developed by the company. The primary objective is to explore and compare various methods for heart rate detection and signal processing, enhancing the understanding of their efficacy and performance. The dataset was collected in 2022 and it is composed of BCG signals and ECG signals and it comprises 20 volunteers signals, for a total test duration of 30 minutes exploiting different positions on the mat and different breathing rates for each position. The investigation integrates four principal methods: Pan-Tompkins algorithm, Independent Component Analysis (ICA), autocorrelation, and Continuous Wavelet Transform (CWT). The Pan-Tompkins algorithm is employed for heart rate detection, while Independent Component Analysis is used as a signal processing technique to better the signal quality. Autocorrelation serves as an additional approach for heart rate detection, and the Continuous Wavelet Transform is engaged in signal processing to have a time- frequency method to investigate on. The combination of these techniques enables a comprehensive exploration of heart rate extraction from BCG signals, facilitating the evaluation of their strengths and limitations. Eventually, the information retrieved from each method are the average heart rate, the heart rate trend over time, the heart rate correlation between BCG and ECG and the HRV parameters correlation. The experimental outcomes reveal a satisfactory overall performance across the methods studied, even though not all the result goals were fully met. Notably, Independent Component Analysis emerges as the optimal algorithm for signal processing, showcasing superior capabilities in enhancing signal quality. On the other hand, the Pan-Tompkins algorithm exhibits the highest accuracy in heart rate detection. Regarding the correlation analysis, the best results were obtaining in methods involving ICA, with a Pearson’s coefficient of 0.957 and a p-value of 0.133. On the other hand, HRV parameters correlation was significantly low. These findings contribute to understand the relationship between heart rate extraction techniques and signal processing methods, highlighting their respective advantages in specific contexts, as well the corresponding correlation between BCG data and ECG, which is considered the gold standard. In conclusion, this thesis offers a comprehensive analysis of heart rate feature extraction from ballistocardiogram signals under optimal conditions. By leveraging the capabilities of the Pan-Tompkins algorithm, Independent Component Analysis, autocorrelation, and the Continuous Wavelet Transform, the research provides valuable insights into their comparative performance. The study's outcomes have implications for the advancement of non-invasive cardiovascular monitoring techniques, underscoring the importance of algorithm selection and signal processing in extracting accurate heart rate information from BCG signals.