Digital Image Processing (3rd Edition)
Digital Image Processing (3rd Edition)
Automatic classification of lymphoma images with transform-based global features
IEEE Transactions on Information Technology in Biomedicine
Multiphase Level Set Model with Local K-means Energy for Histology Image Segmentation
HISB '11 Proceedings of the 2011 IEEE First International Conference on Healthcare Informatics, Imaging and Systems Biology
| Hi-index | 0.00 |
Hodgkin lymphoma (HL) is a special type of B cell lymphoma, arising from germinal center B-cells. Morphological and immunohistochemical features of HL as well as the spatial distribution of malignant cells differ from other lymphoma and cancer types. Sophisticated protocols for immunostaining and the acquisition of high-resolution images become routine in pathological labs. Large and daily growing databases of high-resolution digital images are currently emerging. A systematic tissue image analysis and computer-aided exploration may provide new insights into HL pathology. The automated analysis of high resolution images, however, is a hard task in terms of required computing time and memory. Special concepts and pipelines for analyzing high-resolution images can boost the exploration of image databases. In this paper, we report an analysis of digital color images recorded in high-resolution of HL tissue slides. Applying a protocol of CD30 immunostaining to identify malignant cells, we implement a pipeline to handle and explore image data of stained HL tissue images. To the best of our knowledge, this is the first systematic application of image analysis to HL tissue slides. To illustrate the concept and methods we analyze images of two different HL types, nodular sclerosis and mixed cellularity as the most common forms and reactive lymphoid tissue for comparison. We implemented a pipeline which is adapted to the special requirements of whole slide images of HL tissue and identifies relevant regions that contain malignant cells. Using a preprocessing approach, we separate the relevant tissue region from the background. We assign pixels in the images to one of the six predefined classes: Hematoxylin^+, CD30^+, Nonspecific red, Unstained, Background, and Low intensity, applying a supervised recognition method. Local areas with pixels assigned to the class CD30^+ identify regions of interest. As expected, an increased amount of CD30^+ pixels is a characteristic feature of nodular sclerosis, and the non-lymphoma cases show a characteristically low amount of CD30^+ stain. Images of mixed cellularity samples include cases of high CD30^+ coloring as well as cases of low CD30^+ coloring.