Segmentation of Chromatographic Particles

• Main Authors: Wei-chih Chen, 陳威志
• Other Authors: Chia-ling Tsai
• Format: Others
• Language: en_US
• Online Access: http://ndltd.ncl.edu.tw/handle/25192485708170639372

碩士 === 國立中正大學 === 資訊工程所 === 95 === Chromatography is broadly used in protein analysis and research. Direct intraparticle diffusion analysis (IDA) is a novel approach for the determination of intraparticle protein diffusion coefficients. In order to estimate the diffusion coefficients, the circular chromatographic particles are delineated in the images taken by confocal laser scanning microscopy. With the high image throughput during the experiments, manual segmentation is extremely time-consuming, if not impossible, and the results vary between operators. A system that could automatically and robustly segment the particles and find the centers and radii of particles is needed, and that is the target of this thesis. We use the watershed segmentation to segment the particle region. However, the classical watershed is sensitive to the noise, so it often fails on chromatographic images, which have low signal-to-noise ratio (SNR). We applied an improved anisotropic diffusion filter—anisotropic median-diffusion filter—as the preprocessor of the watershed. This filter is an edge preserving filter that preserves the edges and reduces the impulse noises in an iterative manner. The resulting images with higher SNR are more apt for watershed segmentation. Segmented regions are further processed for extraction of circular particle regions. Circle fitting is performed on regions of reasonable sizes, and the centers and radii are computed for IDA experiments. For each candidate region, the fitting process begins with distance transform to locate the initial center and the radius, and iterates between outlier deletion and least-squares estimation until converged. In this thesis, a total of 67 chromatographic images and 15 artificial images were tested, in which 86.1% of particles in chromatographic images and 94.3% of particles in artificial images were accurately detected. Most of the missing circles were due to the failure of watershed segmentation, and such particles are often not of interest for the IDA analysis. We will improve the accurate rate and get the automatic watershed threshold setting method more accurate in the future work. We also conceive the faster segmentation method for temporal chromatographic images to improve the efficiency of the system.