This title appears in the Scientific Report :
2014
Three-dimensional reconstruction of histological blockface images using ID-encoded markers
Three-dimensional reconstruction of histological blockface images using ID-encoded markers
Introduction:The alignment of serial histological sections of the human brain into an anatomically realistic space requires prior elimination of various distortions, which are inevitably introduced during tissue processing, e.g. cutting and mounting [2]. Blockface images provide largely undistorted...
Saved in:
Personal Name(s): | Schober, Martin (Corresponding author) |
---|---|
Schlömer, Philipp / Hütz, Tim / Cremer, Markus / Mohlberg, Hartmut / Amunts, Katrin / Axer, Markus | |
Contributing Institute: |
Strukturelle und funktionelle Organisation des Gehirns; INM-1 |
Published in: | 2014 |
Imprint: |
2014
|
Conference: | 20th Annual Meeting of the Organization for Human Brain Mapping, Hamburg (Germany), 2014-06-08 - 2014-06-12 |
Document Type: |
Abstract |
Research Program: |
Theory, modelling and simulation Postnatal Development of Cortical Receptors and White Matter Tracts in the Vervet Supercomputing and Modelling for the Human Brain Imaging the Living Brain |
Publikationsportal JuSER |
Introduction:The alignment of serial histological sections of the human brain into an anatomically realistic space requires prior elimination of various distortions, which are inevitably introduced during tissue processing, e.g. cutting and mounting [2]. Blockface images provide largely undistorted images of brain sections. Thus the aligned volume of blockface images represents an important reference to recover the spatial coherence of the non-linearly deformed histological sections. We introduce a robust and efficient method for an automatic 3D reconstruction of blockface images taken from the brain during sectioning (Fig.1).Materials and Methods:The method is based on the use of ID-encoded markers, which have been established in the fields of augmented reality and computer vision to trace camera positions and orientations in real-time. Each marker represents an individual identifying number and encodes a 12-bit number in a 6x6 array of black and white pixels (Fig.1). This pattern guarantees a robust code identification by a “majority vote”, which is important in an environment, where the markers are likely to be partially covered, e.g. by ice crystals or sectioning residues. The detection of the markers was realized by implementation of the ARToolkitPlus library [3]. The basic idea is to extract the coordinates of the same markers in different blockface images and to align them to each other by means of affine transformations. Using only marker matching in the background causes perspective errors in the brain tissue, since the sectioning plane of the brain and the background containing the marker patterns have decisively different distances to the camera lens. Therefore, the median along the depth of the marker-based registered volume was calculated to eliminate outliers. In a next step, the marker-based registered images were aligned slice by slice onto the median volume using a translation transform estimated by the pixel-based image registration algorithm provided by the ITK library [4]. This procedure was applied to different types of blockface images acquired from both paraffin embedded and frozen brains in different types of microtomes with various camera setups and several marker patterns. The example shown here was taken from a formalin-fixed, left human temporal lobe sectioned in a large-scale cryostat microtome, which was further processed for 3D-PLI [1].Results:While the robustness of the marker detection algorithm showed strong dependencies on the marker size and the distance between the markers in relation to the camera resolution, the robustness of the estimation of affine transformations between two sets of coordinates depended on the amount of markers, which is limited by the camera's field of view. Best results were achieved with a marker size of 4mm, a gap of 2mm and a number of about 750 markers. Using this setup, the algorithm detected 95% of all visible markers on average. This step provided pixel-precise results with respect to the marker alignment. In case of using small-scale microtomes for small tissue samples from rodent or monkey brains, for instance, the perspective errors in the cutting plane were most significant due to the close distance to the camera lens. An elimination of perspective outliers in the cutting plane using an alignment to the median of the marker-based aligned images provides an accuracy of less than 4 pixels. Since the algorithm was parallelized, the processing time of 1000 sections could be reduced to one hour. Fig.2 shows the quality of the finally generated blockface volume of the human temporal lobe in reconstructed sagittal and transversal views. |