Computerized Alignment and Deformation Correction of Microscopic Serial Section Images of Nervous Tissue

Abstract: This thesis describes an investigation into digital techniques for image alignment and deformation correction with special reference to serial light and electron microscopic images. These techniques are applied in three-dimensional (3D) reconstruction of nervous tissue, especially the superficial gray matter of the brain - cortex cerebri. Images are based on consecutive two-dimensional (2D) sections, mechanically cut to a thickness of approximately 0.1 - 1.0 µm. Reconstructions comprise from 100 to 5000 sections. The sections are photographed, or captured by video, digitized and fed to a computer.

Since orientation is lost as a result of the sectioning procedure, accurate image alignment is needed. Furthermore, geometrical correction is needed because deformation artefacts due to sectioning are present.

Alignment is carried out by a regional image matching procedure by which two consecutive images are translated and rotated until best match (here, minimal gray scale difference) is found. Sectioning deformation is described by a bilinear model and detected by a matching technique, similar to the alignment procedure. Geometrical correction is carried out by bilinear resampling. On the basis of accurately aligned and deformation corrected images, a stack of 2D images including, for example thousands of brain cells, is created. The geometry and distribution of these cells may now be studied by applying sophisticated computer graphics techniques.

The presented techniques for 3D reconstruction from serial 2D images of the brain and nerve fibres should be of importance for the understanding of normal as well as abnormal brain structure and with applications to biomedical research on, for example, epilepsy and Alzheimer's disease.