Reconstructie van golfvoortplanting in het hart uit mechanische deformatiedata met physics‑informed neurale netwerken

The heart's ability to pump blood relies on tightly coordinated electrical and mechanical processes, and disruptions in these processes can lead to dangerous arrhythmias. Because current clinical imaging of electrical activity is invasive and limited, this thesis explores whether mechanical deformation, which can be captured non‑invasively through echocardiography, could be used to reconstruct underlying electrical activation patterns. To solve this challenging inverse problem, we explore physics‑informed neural networks (PINNs), which blend machine‑learning flexibility with mechanistic cardiac models. In simplified two‑dimensional experiments, the method accurately recovers focal and spiral activation patterns and remains robust under noise, sparse data, and low spatial resolution. The framework is then extended to more realistic three‑dimensional geometries and anisotropic tissue, where it continues to reconstruct activation reliably. In the thesis we also apply PINNs to shear‑wave propagation to estimate tissue stiffness, demonstrating improved velocity‑field reconstruction compared to traditional approaches with seperate smoothing. Validation on synthetic, phantom, and preliminary clinical data highlights both the promise and remaining challenges of the method. Overall, the results show that PINNs can address key inverse problems in cardiac mechanics and may support future development of patient‑specific digital heart twins using fast, inexpensive deformation data.