Machine Learning-Guided Discovery of Copper(I)-Iodide Cluster Scintillators for Efficient X-ray Luminescence Imaging

Abstract

Developing efficient scintillators with environmentally friendly compositions, adaptable band gaps, and robust chemical stability is crucial for modern X-ray radiography. While copper(I)-iodide cluster crystals show promise, the vast design space of inorganic cores and organic ligands poses challenges for conventional approaches. In this study, we present machine learning-guided discovery of copper(I)-iodide cluster scintillators for efficient X-ray luminescence imaging. Our findings reveal that combining base learning models with fused features enhances model generalization, achieving an impressive determination coefficient of 0.88. By leveraging this approach, we obtain a high-performance Cu(I)-I cluster scintillator, named copper iodide-(1-Butyl-1,4-diazabicyclo[2.2.2]octan-1-ium)₂, which exhibit radioluminescence 56 times stronger than that of PbWO₄, and enables a detection limit for X-rays of 19.6 nGy_air s^-1. Furthermore, we demonstrate the versatility of these scintillators by incorporating them as microfillers in the fabrication of flexible composite scintillators for X-ray imaging, achieving a static resolution of 20 lp mm^-1 and demonstrating promising performance for dynamic X-ray imaging.

Keywords: X-ray luminescence imaging; copper(I)-iodide cluster; machine learning; scintillators.