Over the past several years, artificial intelligence has been increasingly applied to imaging assessment in adolescent idiopathic scoliosis (AIS). Automatic Cobb angle measurement is becoming relatively mature1, but studies that move beyond parameter measurement to automate Lenke classification remain limited. Previous studies have shown the feasibility of using computational methods to assist Lenke classification2. The study by Xu et al. examines Lenke classification from the perspective of system implementation and workflow organization, providing an additional contribution to this evolving area. The topic addresses a relevant clinical need and has both practical and academic importance. The main contribution of this study involves the way that the classification process is structured, rather than improvements in algorithmic performance. While prior research has demonstrated that individual radiographic parameters can be measured automatically1, clinical practice requires the integration of multiple measurements within established classification logic to support decision-making. Xu et al. approach Lenke classification as a structured process, systematically organizing its key decision points and linking them into a modular framework. This design enhances the transparency and reproducibility of the classification workflow compared with traditional approaches that rely heavily on clinical experience. Accordingly, the development of artificial intelligence-assisted classification systems may benefit more from careful modeling of the overall decision-making process rather than from isolated optimization of computational steps. Interpretability is another key strength of this work. In spine surgery, barriers to clinical adoption of artificial intelligence increasingly involve concerns about trust rather than raw accuracy. Xu et al. present a white-box model that follows the familiar logic of Lenke classification and aligns closely with established clinical reasoning. This system allows each decision point to be traced back to specific radiographic and anatomical criteria. Such interpretability is critical for meaningful integration of automated tools into routine surgical practice. Several limitations should be acknowledged. First, the findings are derived from data obtained within a limited setting, and external validation across multiple centers has not yet been performed. Given the well-recognized variability in Lenke classification among surgeons with different training backgrounds and levels of clinical experiences, further evaluation in diverse populations is warranted to assess the robustness and generalizability of this system. Additionally, the system is currently intended for research purposes. Future development into clinically applicable tools, such as image-based applications or integrated platform solutions, may facilitate broader use and enable assessment of its performance in real-world settings. Overall, the study by Xu et al. presents a clear and organized approach to automating the traditional 2-dimensional (2D) Lenke classification workflow. Its principal contribution lies in translating a familiar clinical decision process into a structured and interpretable computational system. While the absence of axial rotation and other 3-dimensional (3D) parameters reflects inherent limitations of the classic Lenke framework, the study nevertheless offers a practical example of how automated classification can be implemented in a clinically intelligible manner. Given the recent emergence of 3D classification frameworks, such as the SRS-Lenke-Aubin system3, future classification models will need to incorporate more comprehensive and anatomically faithful 3D information while maintaining interpretability. Three-D models derived from 2D radiographs inherently rely on projection assumptions and shape priors, which limits their ability to characterize vertebral rotation, intervertebral relationships, and true spinal morphology under weight-bearing conditions4. Accordingly, future efforts in 3D classification should prioritize real weight-bearing volumetric spinal imaging rather than simulated 3D reconstructions5. In this context, the work by Xu et al. on automated 2D Lenke classification may be viewed as an important foundational step toward more advanced, clinically relevant decision-support systems. It also provides methodological insights that may inform the future development of 3D classification frameworks based on real weight-bearing imaging data.
Yin et al. (Thu,) studied this question.