LEŠKOVSKÝ ET AL.: REGISTRATION OF HISTOLOGICAL SLICES 1 Annals of the BMVA Vol. 2012, No. 10, pp1–18 (2012) Point-based registration of high-resolution histological slices for navigation purposes in virtual microscopy Peter Leškovský 1 , Alexander Alekseychuk 1 , Adam Stanski 1 , Olaf Hellwich 1 , Karsten Schlüns 2 , Norman Zerbe 2 , Peter Hufnagl 2 1 Computer Vision & Remote Sensing,Technical University Berlin, Berlin, Germany 2 Dept. Digital Pathology and IT, Institute of Pathology, Charité - Universitätsmedizin Berlin, Berlin, Germany Abstract Automatic registration methods for histological images in virtual microscopy are pre- sented. Point-based approaches with outlier filtering by RANSAC and multibody fac- torization are investigated. The algorithms are tested on high-resolution biopsy images of the gastrointestinal tract. Quantitative evaluation is performed at different resolution levels. The results reveal the superiority of Harris feature points over SIFT features, in our setting. At low resolutions or when only subsequent slices are aligned, favorable registration results are obtained using only raw feature matching and Delaunay-based interpolation of the transformation parameters. Outlier filtering is necessary to obtain stable results in more difficult scenarios, i.e. when registering nonconsecutive slices. 1 Introduction Virtual microscopy describes the examination of microscopic structures on digitized images. With last generation optical scanners, taking less than two minutes for the acquisition of standard histological slides (6x2cm 2 ) with 40-fold magnification, the digitization settles an important role in diagnostic pathology. It avoids problems with proper storage of tissue sam- ples and promises easy sharing of histological samples, now in digitized form, among the medical doctors over long distances. It also allows for editing annotations and marking cor- responding regions among several slides. Such placemarks make the navigation over or be- tween histological slides possible. The navigation feature is important during detailed spec- imen diagnosis where examination between several sections is often unavoidable, e.g. when tracking the propagation of malignant tumors. c  2012. The copyright of this document resides with its authors. It may be distributed unchanged freely in print or electronic forms.