Cardiovascular Digital Twin Using Windkessel Model and Neural Parameter Estimation

SIMULATED DATA: VIRTUAL SUBJECTS, NOT PATIENTS. NOT A MEDICAL DEVICE. Can one aortic pressure beat calibrate a cardiovascular digital twin? A three-element Windkessel (peripheral resistance R, compliance C, characteristic impedance Zc) is fitted with the measured flow to all 4,374 virtual subjects of the Pulse Wave Database, a 1-D arterial model. The same parameters are then estimated without flow: the classical diastolic decay plus cardiac output, against a neural network reading the pressure beat alone. The network reaches 1.3 % on R inside the database, but scaling R up and flow down gives exactly the same pressure, so that accuracy is knowledge of the population: with a stroke-volume level never seen the error rises to 21-25 %. What-if answers are checked against pairs of the 1-D model, and a web application (five database paths, SQLite and mongomock tested) presents the results honestly.

Category: Biomedical Engineering

Biomedical Engineering Project BM04 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.