The research team led by Assistant Professor Chao WANG at the Southern University of Science and Technology (SUSTech), in collaboration with the Chinese University of Hong Kong, the Chinese Academy of Sciences, and Hong Kong Baptist University, has achieved a significant breakthrough in X-ray spectro-tomography data processing. The team developed a self-supervised deformation correction algorithm based on implicit neural networks. This work, titled “Data-driven deformation correction in X-ray spectro-tomography with implicit neural networks,” was featured as the cover article in Patterns, a multidisciplinary journal under the Cell Press portfolio.

In the fields of materials science and biology, understanding the 3D interaction between structure and chemical composition at the nanoscale is vital for advancing energy storage technologies. Full-field transmission X-ray microscopy combined with X-ray absorption near-edge structure (TXM-XANES) provides a powerful non-destructive tool for this analysis. However, collecting 2D projections often takes hours, during which mechanical vibrations, optical drift, and electrochemical reactions within the sample cause image misalignment. These deformations introduce artifacts and blur fine features, creating a major bottleneck for the widespread use of X-ray spectro-tomography.

The team introduced the Coordinate Alignment Network (CANet), a data-driven framework that eliminates the need for manual feature labeling or expensive external sensors (Fig. 1). CANet uses implicit neural networks to learn continuous mappings from projection angles or spectral coordinates to affine transformation parameters. Operating on a self-supervised architecture, the model trains directly on the experimental data without needing external datasets. It minimizes loss between aligned projections and reference spectra to achieve high-precision image correction within a unified framework.

Figure 1. Deformation correction methods for X-ray spectral tomography

The algorithm was rigorously validated using simulated data and real imaging of lithium-ion battery cathode materials. While unaligned raw data often suffer from severe motion artifacts that obscure detail, CANet-corrected reconstructions showed a dramatic qualitative improvement. Representative slices revealed that the algorithm suppressed artifacts and restored structural sharpness, making tiny internal cracks in battery particles visible (Fig. 2). This precision in locating fine structures and mapping oxidation state heterogeneity provides essential visual evidence for understanding the physical and chemical degradation mechanisms of batteries.

Figure 2. Practical application of CANet in the alignment of nanotomography and spectroscopic tomography projections of battery cathode particles

The self-supervised CANet algorithm overcomes the technical constraints of traditional microscopy registration by offering an efficient, robust, and automated solution. This advancement not only empowers researchers to design more durable next-generation battery materials but also serves as a general tool for high-resolution 3D and 4D in-situ imaging across various scientific domains.

The research was completed under the joint guidance of Chao WANG, Assistant Professor of the Department of Statistics and Data Science, SUSTech; Jizhou LI, Assistant Professor of the Chinese University of Hong Kong; and Xiqian YU, Researcher of the Institute of Physics, Chinese Academy of Sciences. Ting WANG, a PhD candidate at SUSTech, is the first author of the paper (ranked first among co-first authors), with SUSTech as the first affiliation.

Article Link：https://www.cell.com/patterns/fulltext/S2666-3899(26)00024-3