This paper was accepted on the workshop Deep Generative Fashions for Well being at NeurIPS 2023.
Cardiovascular ailments (CVDs) are a serious international well being concern, making the longitudinal monitoring of cardiovascular biomarkers very important for early prognosis and intervention. A core problem is the inference of cardiac pulse parameters from pulse waves, particularly when acquired from wearable sensors at peripheral physique areas. Conventional machine studying (ML) approaches face hurdles on this context because of the shortage of labeled knowledge, primarily sourced from scientific settings. Concurrently, bodily fashions, just like the whole-body 1D hemodynamics simulators, though informative, wrestle with the inverse drawback and the problems posed by parameter interactions. Latest work has turned to simulation-based inference (SBI) to tell parameter inference by leveraging mannequin simulations. Nonetheless, transferring predictors from simulations to real-world knowledge stays a problem resulting from mannequin misspecifications. Addressing these points, this paper presents a novel hybrid studying method. By fusing a pulse-wave propagation simulator with a data-driven correction mannequin, our methodology goals to mix the rigor of bodily fashions with the flexibleness of ML, providing a promising avenue for efficient cardiovascular biomarker monitoring.