Visualization of Multi-Resolution Reconstructions Using Interactive Stereoscopy

从历史上看，核医学成像一直是使用2D视图进行的。随着……的到来 计算机和断层扫描，3D图像被采集并存储在规则的阵列中 体素。尽管这是2D的逻辑扩展，但将3D对象表示为 2D图像可能不是病变检测的最佳选择。这是因为体素具有 矩形几何图形不能准确地近似成像的几何图形 物体。这种不匹配对核医学研究尤其重要，因为 退化的物理因素迫使图像在由以下部分组成的网格上重建 大立方体素。考虑到病变很小，而且在 从大小到体素，立方体素方法的不足之处显而易见。我们的假设 是通过使用本工作中提出的多分辨率图像表示， 图像重建和呈现比使用 规则网格，从而提高了使用FDG-PET的病变检测能力。我们 选择将图像表示为空间中的一组点(点云)，且不受限制 分配给每个点(节点)的位置和强度，体积由一个集合表示 由节点定义的不重叠的四面体。选择这一点 代表性由两个因素决定。首先，每个图像都有自己的网格， 将被设计成准确和高效地代表每一幅图像。通过使用 不受限制的结点位置，具有任意局部分辨率的几何体 可以对图像体积进行建模。其次，图像的四面体几何图形将 充分利用计算机图形硬件的最新革命性进展 使用高级可视化技术进行立体交互式3D可视化的步骤 成像数据。这里提出的工作将创建一个无体素多分辨率的框架。 核医学中图像的分辨率表示。它包括多学科 层析重建和立体可视化方法的研究进展 使用医学成像和计算机图形学。该工作涉及到设计和创作 用于重建和立体显示基于四面体的图像和 针对人类观察者对病变检测任务的评估。

Historically, nuclear medicine imaging has been done using 2D views. With the advent of computers and tomography, 3D images were acquired and stored in regular arrays of voxels. Although a logical extension from 2D, representation of a 3D object as a stack of 2D images may not be optimal for lesion detection. This is because voxels with rectangular geometry do not accurately approximate the geometry of the imaged objects. This mismatch is especially significant for nuclear medicine studies where degrading physical factors force the images to be reconstructed on grids that consist of large cubic voxels. Considering that the lesions are small and may be comparable in size to voxels, the inadequacy of cubic voxel approach becomes evident. Our hypothesis is that by using the multi-resolution image representation proposed in this work, the images can be reconstructed and presented more accurately and efficiently than using the regular grids, thus resulting in lesion detection improvement using FDG-PET. We chose to represent the image as a set of points in space (point cloud) with unrestricted locations and intensities assigned to each point (node) with volume represented by a set of non-overlapping tetrahedrons defined by the nodes. The selection of this representation was dictated by two factors. First, each image will have its own grid that will be designed to accurately and efficiently represent each image. By using unrestricted node positions, a geometry with arbitrary local resolution that varies across the image volume can be modeled. Second, the tetrahedral geometry of the image will take advantage of recent revolutionary progress in computer graphics hardware in order to use advanced visualization techniques for stereoscopic interactive 3D visualization of the imaging data. The work proposed here will create a framework for voxel-less multi- resolution representation of the image in nuclear medicine. It includes multidisciplinary development of the tomographic reconstruction and stereoscopic visualization methods using medical imaging and computer graphics. The work involves design and creation of software for reconstruction and stereoscopic display of tetrahedron based images and evaluation against human observers for the lesion detection task.