Study On Detection Of Neurons In Zebrafish Embryo Images

Detection of neurons in zebrafish embryo images has great significance for drug discovery research as it's an important means to observe the effect of different drugs. Most of the previous reports on zebrafish image analysis were either based on the manually labeled region of interest (ROI) or visual inspections. Manual analysis and inspection are affected by the problems of inter-rater and intra-rater variations that would significantly impair experimental reproducibility. Further, manual analysis and inspection are very time-consuming and are not scalable to studies involving large datasets. Automatic quantification of neurons in a region of interest (ROI) in zebrafish embryo will improve the efficiency in processing experiment results and establish the foundation for a high throughput screening tool of zebrafish images.In this thesis, an image processing pipeline for detecting and segmenting neurons in zebrafish embryo microscopic images is developed based on the character of zebrafish images. The study consists of three major steps. First, morphological operations are used to segment the zebrafish embryo from the background. Then, a ROI enclosing the neurons is automatically detected taking advantage of a priori information about the location and anatomy of an embryo. Image position calibration, watershed algorithm, adaptive threshold binarization, boundary curve fitting and nearest neighbor interpolation-algorithm are used to automatically extract the region of interest. The length and curvature of the ROI's centerline are calculated. At last, referring to the traditional image labeling algorithm of cell counting, region tracking based algorithm is presented to segment and quantify the valid neurons in ROI, which compares with the circular Hough transform algorithm.A common platform of detecting neurons in zebrafish embryo images is designed and developed using MATLAB based on the image processing pipeline for detecting and segmenting neurons in zebrafish embryo microscopic images. The zebrafish embryo image samples from Brigham and Women's Hospital are used to evaluate the algorithm. The experimental results show that computerized analysis algorithm presented by us has a high accuracy and fast computational speed. The approach can generate accurate results in automatically finding and quantifying neurons when the embryos have different orientations and curvatures, demonstrating the feasibility of using image processing methods to extract phenotypic information from zebrafish embryos.