Radiotherapy, a cornerstone of cancer treatment, is critically limited by tumor radioresistance and off-target toxicity. Lanthanide-based nanomaterials (Ln-NPs) have recently emerged as a versatile and promising class of theranostic radiosensitizers to overcome these hurdles. This review comprehensively outlines the state-of-the-art in Ln-NP-enabled radiotherapy, beginning with their fundamental physicochemical properties and synthesis and then delving into the multi-level mechanisms of radiosensitization, including high-Z element-mediated physical dose amplification, catalytic generation of reactive oxygen species (ROS), and disruption of DNA damage repair pathways. The unique capacity of certain Ln-NPs to serve as MRI contrast agents is highlighted as the foundation for image-guided, dose-painting radiotherapy. We critically summarize the preclinical and clinical progress of representative systems, benchmarking them against other high-Z nanomaterials. Finally, this work discusses the ongoing challenges, such as biocompatibility, targeted delivery, and regulatory hurdles, and envisages future directions, including combinatorial strategies with immunotherapy and the development of personalized nanotheranostic paradigms. Through this synthesis, this review aims to provide a clear roadmap for the continued development and clinical integration of lanthanide nanotheranostics in oncology.

Detecting wind turbine blade cracks in OCT images is crucial to wind power generation safety. However, it is often difficult for existing methods to detect the crack shape in the case of low image contrast, high noise and complex crack shape. This paper proposes a dual-coding network called OCT-DCAFNet, which uses auxiliary features to segment the OCT image cracks of wind turbine blades. We have verified the performance of the network on the OctCrack dataset. The OctCrack dataset uses porcelain cracks as a structural agent because their morphology is similar to that of leaf composite cracks. OCT-DCAFNet contains three independent parallel encoders. A CNN branch processes fine local texture details; a Transformer branch captures global context information; a RetinaNet ResNet50 branch provides target detection prior information to enhance feature representation. Each branch observes the crack image from a different perspective. The two modules integrate the information of these branches. The Transformer Channel Fusion (TCF) module exchanges channel information and applies spatial adaptive fusion to enhance feature complementarity. The adaptive two-way fusion module (ABFM) adopts the two-way cross-attention mechanism and deep separable convolution for further optimization. On the OctCrack dataset, the accuracy rate of OCT-DCAFNet reached 96.60%, the recall rate reached 96.24%, the F1 value reached 96.42%, and the IoU reached 93.09%. These data show that the network can reliably detect cracks under the surface in the actual industrial non-destructive testing environment.