An Image Processing Chain for a Three-Dimensional Reconstruction of Basal Cell Carcinoma.

Histological sections are used regularly for tissue study in clinical surge-ry. Such a restricted two-dimensional view of, e.g. tissue anomalies, can never reveal as much detailed information as that found in the original 3-d specimen. Therefore a 3-d elevation of interesting features would signifi-cantly improve the insight into both spatial extents and local distributions. Actually, a deeper knowledge of its properties is missing. We provide a chain of image processing steps starting from slices and de-scribing the way to a 3-d volume reconstruction of the extracted carcino-ma. Method We use an excised specimen of a basal cell carcinoma embedded in par-affin and consecutively sliced in vertical direction. These slices were then stained with Hematoxylin-Eosin (H&E) and digitized with a Zeiss Mirax Slide Scanner. The image processing chain starts with a first rigid regist-ration (including scaling). Fig. 1. The slides before (left) and after the rigid registration step. After this first registration step the images are processed through a Total Variation Filter for edge-preserving smoothing. The filter uses the estima-ted standard deviation of the noise as only input parameter and is estima-ted from background parts of the image. The purpose of the smoothing is to get a consistent segmentation result as noise and color variations in the images will be blurred to a certain level after ap-plication. The results of the filtering are then segmented using the Fuzzy C-Means Segmentation. We chose this segmentation because it is very li-kely that in Hematoxylin-Eosin-stained slides the segmentation-classes of tumorous and non-tumorous parts overlap in color-space [1]. The Fuzzy C-Means Segmentation can handle this behavior and therefore provided constant good results throughout the whole image stack. Results from the rigid registration step