From radial to aortic pressure: In silico data-driven estimation using an autoencoder

Accurately estimating aortic blood pressure (ABP) waveforms is crucial for cardiovascular health assessment and clinical intervention. However, direct measurement of ABP is challenging owing to its invasiveness and the complex arterial dynamics involved. To address this issue, we introduce a machine learning-based approach utilizing a cross-encoder with an autoencoder–hetero-decoder (AE–HD) architecture. This transfer function is designed to precisely convert radial blood pressure (RBP) waveforms into ABP waveforms. The training phase comprises an auto-encoding step to learn latent RBP representations and a hetero-decoding step to estimate the ABP waveform from the encoded RBP waveform. The proposed model was trained and evaluated on in silico datasets, achieving mean absolute errors of 0.53 ± 0.37, 0.46 ± 0.49, and 0.30 ± 0.28 mmHg for the total waveform, systolic blood pressure, and diastolic blood pressure, respectively, and 2.14 ± 2.25 % for the augmentation index estimation. Although the AE–HD model demonstrated high accuracy in these simulations, its generalizability to real-world clinical settings remains to be established. Further validation using real patient data and diverse clinical scenarios is planned. This research holds potential for improving non-invasive cardiovascular monitoring and enhancing patient care across clinical contexts.