Continuous blood-pressure tracking is a key goal of healthcare AI, yet cuffs are still the norm in clinics and are unusable for this. We present a cuffless model that couples a Physics-Informed Neural Network (PINN) with basic hemodynamic laws to estimate full arterial pressure waveforms, from the aorta to the fingertip, using only an approximate inflow waveform Q(t) and a fingertip photoplethysmography (PPG) and electrocardiogram (ECG) signal. The network contains the Windkessel and 1-D Navier-Stokes models directly in its loss, so it learns realistic wave travel while needing no patient-specific calibration. Trained on records from the MIMIC-III Waveform Database, the network reproduces the expected changes in pulse shape along the arterial path. These results show that combining first-principles physics with deep learning could be a reliable, low-cost blood-pressure sensing.

Physics Informed Neural Network for Continuous and Cuffless Arterial Blood Pressure / Delrio, Federico; Randazzo, Vincenzo; Cirrincione, Giansalvo; Pasero, Eros. - ELETTRONICO. - (2025), pp. 143-148. (Intervento presentato al convegno 2025 IEEE International Conference on Artificial Intelligence for Learning and Optimization (ICoAILO) tenutosi a Bali (Idn) nel 07-09 August 2025) [10.1109/icoailo66760.2025.11156047].

Physics Informed Neural Network for Continuous and Cuffless Arterial Blood Pressure

Delrio, Federico;Randazzo, Vincenzo;Pasero, Eros
2025

Abstract

Continuous blood-pressure tracking is a key goal of healthcare AI, yet cuffs are still the norm in clinics and are unusable for this. We present a cuffless model that couples a Physics-Informed Neural Network (PINN) with basic hemodynamic laws to estimate full arterial pressure waveforms, from the aorta to the fingertip, using only an approximate inflow waveform Q(t) and a fingertip photoplethysmography (PPG) and electrocardiogram (ECG) signal. The network contains the Windkessel and 1-D Navier-Stokes models directly in its loss, so it learns realistic wave travel while needing no patient-specific calibration. Trained on records from the MIMIC-III Waveform Database, the network reproduces the expected changes in pulse shape along the arterial path. These results show that combining first-principles physics with deep learning could be a reliable, low-cost blood-pressure sensing.
2025
979-8-3315-6928-0
File in questo prodotto:
File Dimensione Formato  
2025_Conf_ICoAILO2025_Physics_Informed_Neural_Network_for_Arterial_Blood_Pressure.pdf

accesso riservato

Tipologia: 2a Post-print versione editoriale / Version of Record
Licenza: Non Pubblico - Accesso privato/ristretto
Dimensione 1.71 MB
Formato Adobe PDF
1.71 MB Adobe PDF   Visualizza/Apri   Richiedi una copia
m50516-delrio paper_postPrint.pdf

accesso aperto

Tipologia: 2. Post-print / Author's Accepted Manuscript
Licenza: Pubblico - Tutti i diritti riservati
Dimensione 1.53 MB
Formato Adobe PDF
1.53 MB Adobe PDF Visualizza/Apri
Pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11583/3003672