MoSt-DSA: Modeling Motion and Structural Interactions for Direct Multi-Frame Interpolation in DSA Images

人工智能已成为医学图像分析的重要工具。作为一种先进的脑血管造影技术，数字减影血管造影（DSA）存在一个挑战，即人体受到的辐射剂量与图像数量成正比。通过减少图像并使用人工智能插值，可以显著降低辐射。然而，DSA图像呈现出比自然场景更复杂的运动和结构特征，使得插值更具挑战性。我们提出了MoSt-DSA，这是第一个使用深度学习进行DSA帧插值的工作。与提取不清晰或粗粒度特征的自然场景视频帧插值（VFI）方法不同，我们设计了一个通用模块，以有效的完全卷积方式模拟帧之间的运动和结构上下文交互，通过调整最佳上下文范围并将上下文转换为线性函数来实现。由此受益，MoSt-DSA也是第一个能够在训练和测试期间通过一次前向传递直接实现任意数量的插值和任意时间步长的方法。我们与7个代表性的VFI模型进行了广泛的比较，用于插值1到3帧，MoSt-DSA在470个DSA图像序列（每个序列通常包含152张图像）上展示了强大的结果，平均SSIM超过0.93，平均PSNR超过38（标准偏差分别为0.030和3.6），在准确性、速度、视觉效果和内存使用方面全面实现了最先进的性能。我们的代码可在https://github.com/ZyoungXu/MoSt-DSA上获得。

DSA images present complex motion and structural features, making frame interpolation challenging. The paper aims to propose a deep learning method for DSA frame interpolation and reduce radiation dose to humans.

The paper proposes MoSt-DSA, a deep learning method that models motion and structural context interactions between frames in an efficient full convolution manner by adjusting optimal context range and transforming contexts into linear functions. MoSt-DSA is the first method that directly achieves any number of interpolations at any time steps with just one forward pass during both training and testing.

MoSt-DSA demonstrates robust results across 470 DSA image sequences, with average SSIM over 0.93, average PSNR over 38, comprehensively achieving state-of-the-art performance in accuracy, speed, visual effect, and memory usage. The code is available at https://github.com/ZyoungXu/MoSt-DSA.

Recent related studies include Video Frame Interpolation (VFI) methods for natural scenes, such as DAIN, CAIN, and TOFlow.