The Development Of An Automated Imaging And3D Reconstruction System For Large Scale Serial Tissue Sections

More than ever before, it is now widely recognized that a detailedknowledge of the fine structure of tissues and cells is the foundation ofcomprehending its biological function. On the one hand, observing thestructure of a healthy tissue often provides unexpected details critical forunderstanding its physiological function, while on the other hand, analysisof the structure of a diseased tissue, such as a tumor, provides essentialinformation to clarify the progression of the pathology. In this thesis, acomputer aided automated imaging and3D reconstruction system for largescale assembly and analysis of serial tissue sections was established. Highresolution images of finely sliced samples were acquired automatically andthen reconstructed in3D with this system, enabling a rapid and highlyefficient means by which to obtain objectively justified high resolutionstructural characterizations of a wide range of thick biological tissues.The main work of this thesis consists of the following:First, an automatic imaging platform for large scale serial tissuesections was successfully established. Freely available software wasadapted to enable automated image acquisition of thousands of sequentialserial sections, whether the images were obtained with bright field orfluorescent microscopy. The technical details and the key problemsencountered with this system are herein described.Secondly, with a thorough investigation of the commonly employedmethods for3D reconstruction of serial tissue section images, weimplemented effective procedures specifically designed for fluorescent orbright field images, according to the characteristics of the two kinds of images. In addition, we developed robust techniques, based on phasecorrelation algorithms, to stitch the images together, which is notaddressed by traditional3D reconstruction methods but is especiallyimportant for large scale reconstruction projects. We quantified theeffectiveness of these methods, thereby establishing reliable imagereconstruction schemes for subsequent registration and interpolation steps.Thirdly, a new method for automated segmentation of doubly stainedimmunohistochemical tissue images was developed, which overcomeslimitations of manual approaches as well as of other existing techniques.The method is based on color segmentation and morphological analysis.Compared with a more commonly used method employing colordeconvolution, the procedure developed here results in greatersegmentation accuracy while only requiring one section sample instead ofthree which are necessary for the color deconvolution method.Finally, the effectiveness of the procedures developed here wasdemonstrated with variety of tissues, for both basic science and medicalinvestigations. Among them are thin200nm-thick resin sections ofimmunofluorescently stained mouse brain tissue and human vascularsmooth muscle cell tissue, thicker1μm-thick resin sections from methylblue-magenta stained mouse liver tissue, and5μm thick paraffin sectionsfrom hematoxylin-eosin stained human tongue squamous carcinoma tissue.The automatic, high resolution3D imaging and analysis platformdescribed in this thesis is specifically tailored for large-scale structuralcharacterizations of whole tissues, whether healthy or diseased, enablingthe determination of otherwise unobtainable information about the3Darchitectures of these multicellular systems for both basic science andclinical applications.