%A Kayser, Klaus %A Borkenfeld, Stephan %A Djenouni, Amina %A Kayser, Gian %D 2015 %T Texture and object related image analysis in microscopic images %K Tissue-based diagnosis; content image analysis; MST entropy; Immunohistochemistry; Glycohistochemistry; Segmentation %X Background Tissue based diagnosis or surgical diagnostic pathology undergoes significant changes and focuses on image content analysis in our days. Herein we describe and discuss new approaches of content image analysis and compare their applications, benefits and constraints. Theory Any useful microscopic image contains information that can be evaluated and transferred into a tissue-based diagnosis. A correctly derived diagnosis depends upon the image information and the pathologist’s knowledge, i.e. his ability to recognize and transfer the image content information into clinical application. Thus, image information is related to external “disease” information and “pure” image content information. Application of external image information requires a separation of objects from the background, or segmentation procedures. “Pure” image information is solely pixel based. It can be analyzed using different approaches, such as entropy measure, construction of image primitives and their spatial distribution, or image similarity operations. Our approach uses entropy calculations dependent upon all possible gray value thresholds in combination with syntactic analysis of pixel based image primitives. Implementation Virtual slides underwent the evaluation of regions of interest (ROI) as described previously. ROIs were interactively controlled and subject for application of developed image analysis procedures. The “classic method” of object recognition and syntactic structure analysis is applied too. Trials were performed on antie Her2_new stained, DAB visualized, and glycohistochemically stained, AP visualized slides. The images were measured at magnifications, which correspond to x20, and x40 objectives. Results The algorithm displayed only weak changes of the evaluated gray level based structural entropy (GL-MST)-entropy in the selected ROIs in contrast to the whole image. In addition, significant differences could be obtained when the measures were associated to the clinical impact (diagnosis, fetal developing stage). Conclusions: Images textures and pixel primitives can serve for evaluating “pure” image content information. They do not require segmentation and additional external information for measurement. Interpretation of the measures, i.e. external information can be implemented at a later stage. The described algorithm is probably applicable for image analysis of different fields such as histology, forests, traffic, or can serve for objective image quality evaluation. %U http://www.diagnosticpathology.eu/content/index.php/dpath/article/view/14 %J Diagnostic Pathology %0 Journal Article %R 10.17629/www.diagnosticpathology.eu-2015-1:14 %@ 2364-4893 %8 2015-05-26