Epithelial cell detection in endomicroscopy images of the vocal folds

Abstract

The reasons for a hoarse voice are manifold. Besides structural changes such as additional masses on the vocal folds, changes in the layers of the vocal fold mucus influence the acoustic properties of the voice signal [1]. In our research, we aim to examine this in vivo. One suitable technique for this purpose is the use of micro endoscopes. In contrast to traditional microscopes, the micro endoscopes have a reduced image quality and exhibit strong noise artifacts. Furthermore, images are affected by inhomogeneous illumination. All of the mentioned effects pose a challenge to automatic cell detection and segmentation methods. In this paper, we investigate whether automatic cell detection methods are also suitable for the cells of the epithelium of the vocal folds. Based on band-pass filtering, we could successfully reduce noise and emphasize cell boundaries at the same time. The pass-band was experimentally chosen to emphasize the regular structure of the epithelial cells which can be observed in the frequency domain of the cell image. Subsequently,we applied a watershed segmentation to identify the cell borders. Cell centers were located using a local minima search in the band-pass filtered image. First results indicate thatthe method is able to locate and outline epithelial cells with high accuracy. Future research will focus on the relation between such quantitative measures in cell images to acoustic properties of the voice signal and the mechanical properties of the vocal folds such as the synchrony of their vibration.