Seismocardiography (SCG) captures cardiacinduced chest wall vibrations, offering a non-invasive means of assessing cardiac function. Accurate interpretation of SCG signals relies on high-fidelity recordings, as poor signal quality can lead to misdiagnoses. Traditional signal quality assessment methods often fail to encapsulate the complex waveform dynamics fully. Symmetric Projection Attractor Reconstruction (SPAR) analysis, initially proposed for electrocardiogram, presents a novel approach for visualizing SCG morphology and assessing its quality. However, the sampling frequency (SF) influences SPAR features' effectiveness. This study investigates the impact of SF on SPAR features to validate its suitability for SCG signal quality assessment. Certain commercial wearable devices have fixed SFs, and understanding these effects is key to optimizing acquisition settings. Pearson's correlation coefficient (r) and Cohen's d effect size are employed to evaluate feature consistency and discriminative power across different SFs. The analysis reveals that higher SFs lead to more stable and less randomized features in good-quality signals, resulting in stronger correlations. The highest SF (\mathbf{8 3 3} \mathbf{~ H z}) yields the highest frequency of features with |\boldsymbol{r}|\gt0.7, suggesting improved noise pattern discrimination. Moreover, the SF range of 104 Hz to 416 Hz demonstrates a large effect size for distinguishing between good-And low-quality signals. In contrast, the distinction between good and medium remains moderate to small, being 833 Hz, the only SF that reached a large average effect size. These findings underscore the importance of selecting an appropriate SF based on the desired SCG signal quality assessment. While lower SFs reduce storage and computational demands, they may limit waveform resolution, necessitating upsampling for further analysis. Feature selection techniques could optimize dimensionality without sacrificing classification performance, and merging RGB channel features may improve feature relevance. Future work should explore SPAR parameters, such as time delay and embedding dimension, to enhance waveform representation.
Influence of Sampling Frequency on the Symmetric Projection Attractor Reconstruction
Romano C.;Nusca A.;Ussia G. P.;Schena E.;Massaroni C.
2025-01-01
Abstract
Seismocardiography (SCG) captures cardiacinduced chest wall vibrations, offering a non-invasive means of assessing cardiac function. Accurate interpretation of SCG signals relies on high-fidelity recordings, as poor signal quality can lead to misdiagnoses. Traditional signal quality assessment methods often fail to encapsulate the complex waveform dynamics fully. Symmetric Projection Attractor Reconstruction (SPAR) analysis, initially proposed for electrocardiogram, presents a novel approach for visualizing SCG morphology and assessing its quality. However, the sampling frequency (SF) influences SPAR features' effectiveness. This study investigates the impact of SF on SPAR features to validate its suitability for SCG signal quality assessment. Certain commercial wearable devices have fixed SFs, and understanding these effects is key to optimizing acquisition settings. Pearson's correlation coefficient (r) and Cohen's d effect size are employed to evaluate feature consistency and discriminative power across different SFs. The analysis reveals that higher SFs lead to more stable and less randomized features in good-quality signals, resulting in stronger correlations. The highest SF (\mathbf{8 3 3} \mathbf{~ H z}) yields the highest frequency of features with |\boldsymbol{r}|\gt0.7, suggesting improved noise pattern discrimination. Moreover, the SF range of 104 Hz to 416 Hz demonstrates a large effect size for distinguishing between good-And low-quality signals. In contrast, the distinction between good and medium remains moderate to small, being 833 Hz, the only SF that reached a large average effect size. These findings underscore the importance of selecting an appropriate SF based on the desired SCG signal quality assessment. While lower SFs reduce storage and computational demands, they may limit waveform resolution, necessitating upsampling for further analysis. Feature selection techniques could optimize dimensionality without sacrificing classification performance, and merging RGB channel features may improve feature relevance. Future work should explore SPAR parameters, such as time delay and embedding dimension, to enhance waveform representation.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
