基于特征点的数字病理图像拼接算法研究

发布时间：2019-03-31 07:37

【摘要】：数字病理切片作为远程医师诊断的重要依据之一,其精确性直接决定了医师判断的准确性。由于光学显微镜能够识别病理组织结构上的特征,所以光学显微镜成像技术已经被广泛的用在医学应用领域。为了达到诊断的目的,最终经医生诊断的病理图像应该是病理组织的整体图像,并具有较高的分辨率,且能提供较丰富的关于病理组织的细节信息。目前,最新的数字化的病理扫描系统所能处理的超高分辨率病理组织图像的像素数可以超过400M。但是由于常规图像采集传感器的像素数的限制,我们无法利用常规的图像采集阵列实现一次性扫描如此高像素的图像。即使是使用超高像素的图像采集阵列,不仅传感器价格昂贵,与其配套使用的光学系统制作要求也很高,其成本是我们无法承受的。一个现实可行的解决方法就是采用当前主流技术的图像采集系统,使用高倍率、小视野,配合步进电机驱动的2D可移动精密承载台,对一个病理组织玻片的有效面积内的组织进行边界重叠的多次扫描采集,然后利用快速精确的图像拼接算法把多次采集到的各个子图拼接成一个超高分辨率的完整图片。基于以上背景,本文的主要工作包含两部分,一是设计了一种光学显微镜下的病理组织图像采集设备,二是研究不同的特征点提取算法,改进得到一种快速有效的图像拼接算法来形成整个病理组织样本的全景图像。数字病理图像通过我们的显微镜图像采集系统获得,这些病理图像只是整个病理组织的一小部分,并且具有较高的分辨率,除此之外,这些病理组织图像在物理位置上相邻,并具有一定大小的重叠区域,以免丢失拼接时的边缘细节信息。所有通过图像采集系统获取的图像利用我们改进的快速拼接算法来形成整个病理组织样本的全景图。该系统技术适用于临床病理研究领域。通过对我们设计的病理图像采集设备获取的图像,对改进的拼接方法的可行性和表现与现有的方法以及最新的特征点提取算法在处理临床病理图像方面进行了对比,我们的方法是高效与精确的,使诊断医生可以快速准确地做出判断。
[Abstract]:Digital pathological section is one of the important bases for remote diagnosis, and its accuracy directly determines the accuracy of physician's judgment. Optical microscope imaging technology has been widely used in medical applications because of its ability to identify pathological and structural features. In order to achieve the purpose of diagnosis, the final pathological image diagnosed by doctors should be the whole image of pathological tissue, with high resolution, and can provide more detailed information about pathological tissue. At present, the up-to-date digital pathological scanning system can process ultra-high resolution pathological tissue images with a pixel number of more than 400m. However, due to the limitation of the pixel number of the conventional image acquisition sensor, we can not use the conventional image acquisition array to scan such a high pixel image at one time. Even if we use ultra-high pixel image acquisition array, not only the sensor is expensive, but also the requirement of making optical system is very high. The cost of the sensor is too high for us to bear. A practical solution is to use the current mainstream technology of image acquisition system, using high-rate, small field of view, combined with stepping motor driven 2D movable precision bearing table, The tissue in the effective area of a pathological tissue slide is collected by multiple scanning with overlapping boundaries, and then each sub-image collected many times is stitched together into an ultra-high resolution image using a fast and accurate image splicing algorithm. Based on the above background, the main work of this paper consists of two parts: one is to design a kind of image acquisition equipment for pathological tissue under optical microscope, the other is to study different feature point extraction algorithms. A fast and effective image mosaic algorithm is improved to form a panoramic image of the whole pathological tissue sample. Digital pathological images are obtained through our microscope image acquisition system, these pathological images are only a small part of the entire pathological tissue, and have a high resolution, in addition, these pathological tissue images in the physical position adjacent to each other, And has a certain size of overlapping areas, so as to avoid losing the details of the edge of the splicing information. All the images obtained by the image acquisition system use our improved fast stitching algorithm to form a panoramic picture of the whole pathological tissue sample. This system is suitable for clinical pathological research. This paper compares the feasibility and performance of the improved splicing method with the existing methods and the latest feature point extraction algorithm in the processing of clinical pathological images by comparing the images obtained by the pathological image acquisition equipment designed by us, and comparing the feasibility and performance of the improved splicing method with the existing methods and the latest feature point extraction algorithm. Our approach is efficient and accurate, allowing diagnostic doctors to make decisions quickly and accurately.
【学位授予单位】：山东师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R-05;TP391.41

【参考文献】