Curating dataset for AI -based Stiffness Estimation in MR Elastography Using Finite Element Modeling and Polynomial Curve Fitting

Motivation: The estimation of stiffness in Magnetic Resonance Elastography (MRE) yields erroneous estimates, due to inability of current algorithms to correctly estimate propagating wavelengths. Goal(s): How to curate the dataset, if AI is to be employed as an algorithm to estimate the stiffness values in MRE. Approach: We designed a pipeline to curate a supervised learning dataset using Finite Element Modelling, and Polynomial Curve Fitting. Results: The dataset curated by our pipeline was then used to estimate the stiffness values using supervised learning, with U-Net as our model's architecture; the model's performance was evaluated on some human liver geometries. Impact: Stiffness of the soft tissues is an important biomarker for detecting various pathological states. This method enhances the MRE by enabling the precise tissue stiffness estimation, advancing non-invasive diagnostics of diseases like fibrosis, and cancer.