Region of Interest Queries in CT Scans
Region of Interest Queries in CT Scans
复制标题
CT 扫描中的感兴趣区域查询
DOI：
--
复制
发表时间：
2011
期刊：
影响因子：
--
通讯作者：
Marisa Thoma
中科院分区：
文献类型：
--
作者：
A. Cavallaro;Franz Graf;H. Kriegel;Matthias Schubert;Marisa Thoma
医学图像库包含非常大量的计算机断层扫描(CT)。当查询特定的CT扫描时，用户通常对整个扫描不感兴趣，而是对特定的感兴趣区域(ROI)感兴趣。不幸的是，根据扫描坐标来指定ROI通常不是一个选项，因为ROI通常被指定为w.r.t。扫描内容，例如另一扫描中的示例性区域。因此，系统通常检索完整的扫描，并且用户必须手动导航到ROI。除了导航时间外，加载和传输扫描的不相关部分还有很大的开销。
在本文中，我们提出了一种回答ROI查询的方法，该查询由一个示例ROI在另一次扫描中指定。我们新方法的一个重要特征是，在查询处理之前不需要对查询或结果扫描进行注释。由于我们的方法是基于图像相似度的，因此具有很强的灵活性。扫描区域的大小和位置。为了回答ROI查询，我们的新方法使用了基于实例的回归和内插技术，将扫描的切片映射到人体的身高模型。此外，我们还提出了一种高效的结果扫描搜索算法，以获得高精度的ROI。在实验评估中，我们在2526个CT扫描的数据库上检验了新方法的预测精度和节省的I/O代价。
Medical image repositories contain very large amounts of computer tomography (CT) scans. When querying a particular CT scan, the user is often not interested in the complete scan but in a certain region of interest (ROI). Unfortunately, specifying the ROI in terms of scan coordinates is usually not an option because an ROI is usually specified w.r.t. the scan content, e.g. an example region in another scan. Thus, the system usually retrieves the complete scan and the user has to navigate to the ROI manually. In addition to the time to navigate, there is a large overhead for loading and transferring the irrelevant parts of the scan.
In this paper, we propose a method for answering ROI queries which are specified by an example ROI in another scan. An important feature of our new approach is that it is not necessary to annotate the query or the result scan before query processing. Since our method is based on image similarity, it is very flexible w.r.t. the size and the position of the scanned region. To answer ROI queries, our new method employs instance-based regression in combination with interpolation techniques for mapping the slices of a scan to a height model of the human body. Furthermore, we propose an efficient search algorithm on the result scan for retrieving the ROI with high accuracy. In the experimental evaluation, we examine the prediction accuracy and the saved I/O costs of our new method on a repository of 2 526 CT scans.