Accelerated discovery of MRI-compatible β-Ti alloys via materials database integration and machine learning

Data-driven machine learning approaches facilitate the rapid screening and evaluation of the electromagnetic properties of magnetic resonance imaging (MRI) compatible alloys across a broad compositional space. In this work, we establish a closed-loop interpretable machine learning workflow, combined with a dynamically updated materials database, to enable the rapid evaluation of the magnetic susceptibility and electrical conductivity of MRI-compatible β-Ti alloys. A cross-system materials database of MRI-compatible alloys was constructed by integrating electromagnetic data from Ti-, Zr-, Cu-, and Au-based alloys. Alloy features that may reflect the compatibility between parent and product phases, lattice distortion, and the free energy in the alloy are considered in the model. The established models were then used to evaluate the magnetic susceptibility and electrical conductivity of several new and known alloys, with subsequent optimization based on the dynamic database. Furthermore, the Shapley Additive exPlanations (SHAP) method was employed to interpret the models, revealing that shear modulus mismatch and mixing entropy are the most critical features in the prediction of magnetic susceptibility and electrical conductivity, respectively. These results provide a data-driven strategy for the evaluation and fast screening of MRI-compatible β-Ti alloys, with a deep understanding of MRI-compatibility from the perspective of physicochemical mechanisms.