Mechanism Of Molecular Hyperspectral Imaging And Its Applications In Blood Cells Analysis

The technique of hyperspectral imaging, characterized by acquiring I-dimension spectral information and2-dimensions spatial information for each pixel, was initially applied in remote sensing in1980's, which brings remote sensing into an completely new period——hyperspectral remote sensing. Its application fields broadly include precision agriculture, forestry, water detection, atmospheric pollution detection, biological environment, and so on. In this paper, we combined this technique and microscopy to set up a molecular hyperspectral imaging system (MHIS), and then explore its applications in biomedicine field.The main contributions of this thesis are summarized as follows:1. The accomplishment of Molecular hyperspectral imaging system (MHIS), as platform for our research, includes two aspects of work:hardware and software. For hardware, AOTF,CCD,microscope,DAQ card are respectively collected with computer by serial parallel interface of that. As for software, C++Builder tool is applied to make an integrated software interface, which acts as a controller for AOTF and data acquisition.2. Based on the principle of molecular hyperspectral imaging system, some key technologies had been elaborated in detail, including AOTF, data acquisition system and so on. In addition, the specifications of molecular hyperspectral imaging system had also been introduced.3. The system is susceptible to be affected by spectral response,light path. Therefore, the earliest acquired date must be preprocessed before it is used in analysis. In the paper, we proposed a method for data preprocessing based on blank slide, and the impact of lamp and spectrum response can be eliminated after such preprocessing. The experiment results show that these spectra for different cells are much more comparable compared with those with no preprocessing.4. About characteristic spectrum acquisition, ROI:Regions of interest and COI: Components of interest were comparably provided, and COI was proved to be more preferable by spectral distance based on curve entropy (SDE). The algorithms, binary coding, spectral angle mapper, and spectrl information divergence, were demonstrated to segment white blood cells, and SID was proved to be more effective in our system. The SID segmentation method based on molecular hyperspectral imaging system is good enough to segment white blood cells well, which cast a light on blood analysis. After the segmentation stage, cell characteristics could be extracted accurately including nucleus area, cytoplasm area as well as spectral characteristics, all of them play an important role in blood cells automatic recognition.