Analysis of Cells using Image Data from Sequential Immunofluorescence Staining Experiments Carolina W¨ ahlby , Fredrik Erlandsson , Joakim Lindblad , Anders Zetterberg , and Ewert Bengtsson Centre for Image Analysis at Uppsala University, Uppsala, Sweden Department of Oncology-Pathology at the Karolinska Institute, Stockholm, Sweden ABSTRACT Sequential immunofluorescence staining is a novel method for visualization of large numbers of antigens in single cells in tissue sections. Visualization of large numbers of antigens is often desirable both for quantitative studies and to explore spatial relationships of functional significance. Sequential staining, meaning repeated application and removal of fluorescent markers followed by repeated digital image microscopy, results in large amounts of image data. This paper presents a method for efficient, semi-automatic analysis of image data followed by classification of individual cells as positive, weakly positive, neutral, weakly negative or negative for each of the sequentially applied stains. The method consists of a fully automatic 3D image registration step, semi-automatic 2D and 3D image segmentation, fully automatic extraction of image data as well as classification of cells based on the data histogram using kernel density estimates and fuzzy clustering. Keywords: image registration, sequential immunofluorescence staining, segmentation, classification 1. INTRODUCTION By combining immunohistochemistry with digital image analysis, properties of specifically stained cells in culture or tissue sections can be quantified. The applications include a wide range of fields such as histology and cytology, where quantification of morphological, biochemical, and molecular changes are of great interest. Biological systems are controlled by a very large number of interacting molecules of different kinds. The number of different antigens that can be observed simultaneously using a fluorescence microscope is limited by the specificity of the immunohistochemical detection system, the spectra of the fluorochromes, and the accuracy of the bandpass filters separating the excitation and emission wavelengths. The basic requirement for visualization of a given antigen, i.e. a protein, a protein complex, or other macro molecules, is the availability of a relevant antibody reagent that, when used as a specific probe on the test preparation, can be identified by direct or indirect labeling with a discernible signal. Many different staining systems producing detectable signals at the location of a specific antibody have been developed since the early days of immunohistochemistry [3]. Systems based on fluorescence are most commonly used for quantitative antigen analysis [4]. Fluorescent labels (introduced directly onto a primary antibody, or on a secondary antibody) in combination with narrow band pass filters for limitation of fluorochrome excitation and emission opens the possibility to quantitatively image several antigens in parallel in a single multiply stained tissue section. Double and triple stains have been used to observe as many as three different antigens simultaneously [6]. Many problems occur when a large number of antigens are stained at the same time. The different primary and secondary antibodies can interact and ’bleed-through’ of contrasting emission colors as well as autofluorescence of the stained tissue can disturb the obtained images. In sequential immunofluorescence staining the fluorescence label and primary antibodies are removed after imaging the stained tissue section. This is done without affecting the antigens, and a new set of specific stains can be applied and imaged. By staining, imaging, and washing repeatedly and using digital image analysis, the number of different antigens that can be imaged and quantitatively analyzed simultaneously in single cells in tissue sections increases. The fluorescence present before any stain is applied, and the stain that is left after each washing step is also imaged. Correspondance to Carolina W¨ ahlby, Centre for Image Analysis, L¨ agerhyddsv 17, 75237 Uppsala, Sweden, e-mail: carolina@cb.uu.se