Tumor and organ displacement and deformation due to respiratory motion in radiation therapy are key factors affecting treatment accuracy. Although 4DCT imaging technology can dynamically capture organ motion, its long scanning time and complex operation process limit the application of clinical image guidance. Although current artificial intelligence-based dynamic medical image generation methods have made progress, they still face challenges in the field of 4DCT generation: effective extraction of dynamic anatomical structures and texture evolution of multiple organs; accurate modeling of long timeseries dependencies. To this end, this paper proposes an innovative knowledge-guided dynamic-static feature synergy network (KSPNet). The network has two core technologies: first, the multiscale bi-directional voxel flow attention module is used to efficiently capture the temporal motion information while accurately characterizing the dynamic evolution of anatomical structures and textures, thus realizing the synergy of dynamicstatic features; second, the motion knowledge-guided temporal correction unit is designed to efficiently optimize the modeling of the long-time sequence dependency relationships. The experimental results show that KSPNet performs well in the key indexes of SSIM, PSNR and MS-SSIM, and can accurately generate the lung deformation process during the complete respiratory cycle, and can generate high-quality unobserved time-phase images with only a small amount of input from the 4DCT images, which significantly reduces the dependence on multiple timephase scans.

Medical image synthesis is critical for data augmentation and pathological analysis. While Generative Adversarial Networks (GANs) have been widely used, they suffer from limitations in diversity, stability, and adaptability to few-shot scenarios. Denoising Diffusion Probabilistic Models (DDPMs) have recently shown superior generation quality and stability, but their reliance on large-scale labeled data and high computational costs hinders real-world medical applications. In this paper, we propose a domain-adaptive generative framework tailored for medical image synthesis under limited supervision. Combining domain adaptation with a generative base model, we realize a fast migration from natural to medical image generation. A dual-path training strategy is introduced to jointly optimize the denoising process and encode task-specific conditions into a shared embedding space, allowing the model to focus on clinically relevant features with minimal supervision. Additionally, we propose a data-driven conditional prompt optimization mechanism that guides the model to generate anatomically precise structures, enhancing both fidelity and diversity. Experiments on a multi-modal MRI glioma dataset demonstrate that our method significantly outperforms baseline models in few-shot settings. Our approach provides a novel perspective on diffusion based medical image generation and shows strong potential for downstream clinical tasks.

The transformer has been introduced in the hyperspectral image (HSI) classification, demonstrating outstanding capability in capturing global features compared to the convolutional neural network (CNN). However, the small-sample-size problem poses a significant challenge in practical HSI classification, especially in training the transformer. To tackle this issue, cross-domain transfer learning is adopted as a practical solution, which transfers the information from a source domain with abundant labeled samples to a target domain with limited labeled samples. This article proposes a novel transfer learning method for heterogeneous cross-domain HSI classification called cross-domain discriminative vision transformer (CD-DViT). This algorithm primarily contains three key contributions. First, source samples are mapped to the target domain through an encoder-decoder architecture, and the mapped source samples can be used to train the target classifier. Second, the cross-attention mechanism is utilized to construct two blocks for achieving the domainwise and classwise feature alignments (FAs), respectively. Specifically, the combination of the cross-attention mechanism with the domain discriminator aims to learn domain-invariant features, thereby facilitating domainwise alignment and alleviating domain shift. Third, knowledge distillation (KD) is adopted to learn more information from the target domain and assist in classifying target samples. Our experiments on three real-world cross-domain HSI datasets demonstrate the effectiveness of the proposed approach.