Accurate and Fast 3D Image Reconstruction in Fluorescence Microscopy and Automatic labeling of 3D tissue structures

点击翻译按钮获取中文摘要

This project has been aiming to evaluate and improve the new three dimensional reconstruction algorithm developed for Fluorescence Microscopy that has been proven effective for artificially generated ideal data in our previous project. In order to evaluate the effectiveness of the algorithm for actual data observed from tissue structures, we corrected actual data using a fluorescence microscope and have conducted comprehensive evaluation using the data under different conditions by changing noise level and a coupled of optical parameters. Through this evaluation, we found that the new algorithm is sensitive to PSF (Point Spread Function) parameters and, affected by non-transparent parts that contradicts the ideal situation that the deconvolution-based reconstruction theory assumes. We have spent the most of our time to solve this problem and concluded that reconstructed results would not be improved as far as using deconvolution-based-algorithm and it is necessary to develop a method that would not assume transparency for the objects. After the evaluation and adding some improvement, we also tried to develop matching algorithms for three dimensional structures between actual tissue structures observed by Fluorescence Microscopy and artificially synthesized three dimensional structures. We have examined an initial three dimensional shape modeled by a polyhedron would converge to the target shape by iteration procedure using a neural network specifically designed for the purpose.