Seismocardiography is a potent non-invasive cardiovascular monitoring technique whose widespread adoption is currently limited in ambulatory settings due to its susceptibility to corruption from environmental noise. In the absence of a clean concurrently collected electrocardiogram (ECG) signal as a heartbeat reference, template matching paired with windowing methods can serve as a useful method by which to assess seismocardiogram (SCG) signal quality. However, windowing methods can introduce a time-shift in the segmentation of the SCG beats as compared to a template due to persistently adapting heart rate. In this study, we assess the performance of a state-of-the-art SCG signal quality assessment algorithm, dynamic time feature matching (DTFM), in ranking SCG beats by signal-to-noise ratio when introducing an artificial timedelay. We compare this performance against that of a novel methodology based on topological data analysis (TDA) using persistence diagrams. We found no significant difference$(p>0.05)$in ranking performance between topological data analysis (TDA) and dynamic time feature matching (DTFM) when SCG beats were segmented by true R-peak locations. However, we found that TDA significantly outperformed DTFM$(p<0.001)$when SCG beats were segmented 100, 200, or 300 ms earlier than the R-peak locations. These results suggest the potential promise of TDAbased methods for robust ECG-free SCG signal quality analysis. These advancements may facilitate the analysis of longitudinal SCG data taken in out-of-clinic settings in situations where ECG monitoring is not viable.

Accurate, continuous monitoring of cardiac output (CO) is crucial for effective resuscitation management in hemorrhagic trauma, yet current gold-standard methods are invasive and impractical in field settings. This study introduces a fully non-invasive and wearable sensing-based approach utilizing electrocardiography (ECG), seismocardiography (SCG), and photoplethysmography (PPG) signals, integrated with machine learning algorithms, to enable stroke volume (SV) and CO estimation without requiring baseline calibration or normalization. This critical feature makes the model especially suitable for casualty care scenarios where baseline measurements are often unavailable. The proposed methodology was evaluated on a porcine model$(\mathrm{n}=6)$subjected to controlled hemorrhage and resuscitation protocols. Clinically-validated cardiovascular features were used as inputs for regression models, including linear, ridge, LASSO, random forest, and XGBoost regressors. Among these, the LASSO demonstrated the best performance, achieving a high correlation$(R=0.79)$and a mean absolute percentage error (MAPE) of 14.31%, well within clinically-acceptable limits for non-invasive CO monitoring. The framework reliably tracked SV trends crucial for clinical decision-making during resuscitation scenarios. This work highlights the potential for intelligent, noninvasive CO monitoring systems to improve clinical and trauma care outcomes.

Short-term physiological forecasting holds promise for applications requiring capture of rapid changes in physiological signals. Modeling these transient dynamics may enable the quantification of subtle physiological changes that can be critical in many real-time or sensitive contexts, such as assessing acute stress or closed-loop resuscitation. In this paper, we demonstrate the effectiveness of a time-varying autoregressive Kalman filterbased framework for short-term forecasting of physiological features. To eliminate the need for manual hyperparameter tuning, we integrate an adaptive mechanism that dynamically estimates the process and measurement noise parameters of the Kalman filter. Our results demonstrate that the proposed method outperforms baseline models by approximately 1 ms in predicting the next heartbeat's pre-ejection period and by$2-3 \text{ms}$in predicting the next heartbeat's left ventricular ejection period. Moreover, we demonstrate that it can achieve this improved performance without hyperparameter tuning. This work provides a robust forecasting framework for tracking physiological features and generating new ones to capture short-term physiological variations.

Video frame interpolation (VFI) is a fundamental vision task that aims to synthesize several frames between two consecutive original video images. Most algorithms aim to accomplish VFI by using only keyframes, which is an ill-posed problem since the keyframes usually do not yield any accurate precision about the trajectories of the objects in the scene. On the other hand, event-based cameras provide more precise information between the keyframes of a video. Some recent state-of-the-art event-based methods approach this problem by utilizing event data for better optical flow estimation to interpolate for video frame by warping. Nonetheless, those methods heavily suffer from the ghosting effect. On the other hand, some of kernel-based VFI methods that only use frames as input, have shown that deformable convolutions, when backed up with transformers, can be a reliable way of dealing with long-range dependencies. We propose event-based video frame interpolation with attention (E-VFIA), as a lightweight kernelbased method. E-VFIA fuses event information with standard video frames by deformable convolutions to generate high quality interpolated frames. The proposed method represents events with high temporal resolution and uses a multi-head selfattention mechanism to better encode event-based information, while being less vulnerable to blurring and ghosting artifacts; thus, generating crispier frames. The simulation results show that the proposed technique outperforms current state-of-the-art methods (both frame and event-based) with a significantly smaller model size. Multimedia material: The code is available at https://github.com/ahmetakman/E-VFIA