Reconstruction And Visualization Of Neuronal Morphology For Light Microscope Images

That is one of the important methods for neuroscience research using microscopy to observe and collect images of brain tissue, reconstruct neuronal morphology, to help understand and reveal the mechanism of brain function and behavior. With the progress of experimental methods, especially the significant development of the light microscopy (LM) imaging technique, the type and capacity of neuronal morphology images obtained are increasing rapidly. However, for the large scale LM images analysis, fast and reliable neuronal morphology reconstruction and interactive visualization, is still an open problem.This dissertation studies neuronal morphology reconstruction and visualization method for LM images. The specific content is as follows:In order to detect the location and reconstruct surface of soma (neuronal cell body) in LM image, a shift ray-burst sampling algorithm is proposed and used in combination with a multi-scale image filtering and distance transform algorithm. Several sets of benchmark test data from open source projects are used to evaluate the reliability and performance of the proposed algorithm. The results show that it can accurately detect somas with various densities, sizes and shapes, and reconstruct object close to the sphere. The speed of the algorithm itself is very high, the method of time consumption is mainly on the multi-scale filtering. The proposed method is applied to Micro-Optical Sectioning Tomography (MOST) dataset, to reconstruct somas in local areas, and provide initial points for neurite tracing.In order to trace the trajectory and reconstruct tree-like structure of neurite in LM image, a prediction-correction algorithm is proposed and used alone or in combination with certain image filtering algorithms. Several sets of synthetic data and real LM images are used to evaluate the reliability and performance of the proposed algorithm. The results show that it can trace continuous neurites accurately, and the recall and precision are fairly high. The algorithm running in automatic or semi-automatic mode, can reach the speed of interactive. In general, it achieves a reasonable balance between fast speed and acceptable accuracy, which is attractive for rapid reconstruction of neurites and other tree-like structures. The proposed method is applied to MOST dataset, to reconstruct single neuron and local neural circuits, and applied to angiogram images, to trace and reconstruct vascular structures.In order to visualize large capacity LM images volume data with high-quality, the graphics processing unit accelerated ray-casting is implemented and volume data bricking strategy is adopted. It is integrated with the soma detection and neurite tracing algorithms, to developed a neuronal morphology reconstruction and visualization software named flNeuronSight (open sources). The performance and usability of the proposed algorithm and software are evaluated in application environment. The results show that the current implementation can render volume data more or less of system memory capacity or reconstructed structure containing Hundreds of thousands of nodes neurites accurately at interactive frame rate.The digital reconstruction and visualization of large scale neuronal morphologies is a very complex task, which is a severe challenge for both computing hardware and software design. Various algorithms used in combination should be the direction to solve the problem. Studies in this dissertation has add a set of alternative methods for this field.