Image Reconstruction of the Corneal Subbasal Nerve Plexus with Extended Field of View from Focus Image Stacks of a Confocal Laser Scanning Microscope

Background: Confocal laser scanning microscopy (CLSM) allows the in vivo analysis of nerve structures of the human cornea. In this way, pathological alterations of the peripheral nervous system that also affect the corneal subbasal nerve plexus (SNP) can be diagnosed non-invasively and possibly earlier than with other methods. The field of view of in vivo CLSM images of the cornea (ca. 0.4 X 0.4 mm(2)) is not sufficient for a reliable assessment. Two phenomena make the image assessment difficult: the presence of ridgelike tissue deformations in the neighbourhood of the SNP and image distortions that are induced by involuntary and unavoidable eye movements during image acquisition. This paper presents an image processing method for generating undistorted images of the SNP with an extended field of view. Methods: The presented method has been tested on five volunteers. Eight focus image stacks have been taken and processed from each subject using a Heidelberg Retina Tomograph with Rostock Cornea Module (HRT). An image registration scheme specifically adapted to the image acquisition system corrects the non-linear motion-induced image distortions and reconstructs a volume from each focus image stack. The epithelial basal boundary surface including the SNP appears as a distinctive hyper-reflective layer inside the reconstructed volume. Extracting this continuous layer generates a depth map and finally a two-dimensional image of the SNP. A final fusion step of the single reconstructed SNP images leads to laterally extended images. Results: Out of 40 focus image stacks, 34 have been fully processed into two-dimensional SNP reconstruction images. Six focus image stacks could not be transformed into volumes because of extremely fast eye movements during the image acquisition that prevented the complete image registration of the stacks. The 34 SNP reconstruction images depict an average area of 94.7% (+/- 6.2%) with respect to the field of view of a single HRT image. The final fusion of the reconstructed images resulted in an average increase of the image area by a factor of 2.6 (ranging from 2.2 to 3.1). Conclusion: The presented image processing algorithms are capable of correcting the motion-induced image distortions and of generating larger two-dimensional images of the SNP even in presence of severe tissue deformations. These images provide the basis for a more reliable assessment of the corneal nerve fibres.