Research On The Automated Image Analysis Of Cytokinesis-blocked Micronucleus And Its Realization

Research on the automated image analysis of cytokinesis-blocked micronucleus and its realization[Background]The cytokinesis-blocked micronucleus (CBMN) test in human peripheral blood lymphocytes (HPBL) is an extensively used method for radiation biological dose estimation, health assessment for radiation workers, biomonitoring genetic toxicology, and screening of carcinogen or mutagenicity. In particular, the CBMN test can serve as a biological dosimeter after radiation exposure to estimate the radiation dose between 0.25 to 5Gy with the advantages of economic, simple, and accuracy. However, there are large numbers of micronucleus samples to be analyzed by visual scoring of micronucleus (MN) which requires a trained individual to detect and count at least 1000 binucleated cells (BNC) under the microscope because of statistical reasons, becoming a tedious, subjective, time consuming and error-prone task. All of these have restricted the widely application of CBMN test, which can not meet the urgent needs of health assessment for large numbers of radiation workers or radiation dose estimation for victims after radiation accidence.Compared with the flow cytometry and the laser scanning cytometry which can not be used to analyze the CBMN, the image processing method has the potential to automatically score the BNC and MN in CBMN images efficiently. The automated image system of CBMN has not been implemented thoroughly until now, because it involves many key technologies which are difficult to be accomplished. Several automated analysis systems of CBMN image have already been presented from different external research groups, but a common limitation of these systems is the lower scoring ability and recognition accuracy of BNC and MN which may restrict the statistical reliability of the results. There are no interrelated reports on the automated image analysis internal up to now, except some researcher's desire.[Purpose]To meet the needs of health assessment for large numbers of radiation workers and radiation dose estimation for victims after radiation accidence, we will resolve all the key technologies and difficulties involved in the automated analysis of CBMN images firstly, and then develop an automated image system of CBMN with high accuracy, recognition ability and complete automation. By this image system, the CBMN test can be automated analyzed; the coherence and standardization of scoring criterion can be realized; and the workers can also be extricated from the tedious, time consuming task.[Materials and Methods]1. We firstly adapted the slide preparation protocol to obtain an optimal cell density and dispersion which is important for image analysis, because it is important for image analysis to standardize the slide preparation as much as possible to be able to analyze the input image in an unambiguous and consistent way and to obtain an optimal reproducibility.2. The zeiss-AX10 microscope (Objective 20×) and the Metafer4 system in our laboratory would be utilized to collect the CBMN images with 1280×1024 resolution, by which the whole slide can be converted into 2117 sheets of grey-level images within about 8 minutes.3. The following key techniques should be implemented:(1) Preprocessing the input CBMN images to improve the image quality; (2) Segmenting the input CBMN images to acquire the cell regions, nuclei and micronuclei regions respectively; (3) Classifying the targets in CBMN images into contaminated regions, isolated cells and overlapping regions; (4) Automatic separating the clusters in CBMN images; (5) Automatic scoring of the BNC; and (6) Automatic scoring of the MN, etc.4. We utilize the MATLAB7.1 platform and its graphical user interface (GUI) to accomplish the whole software system including the following modules:(1) Reading of large numbers of CBMN images; (2) Pre-deleting of blank CBMN images; (3) Automatic scoring of BNC and MN; (4) Revisable displaying of multiple BNC images based on MATLAB and HTML hybrid programming.5. To validate the performances of the automated image analysis system, HPBL were irradiated with different doses of 60Co gamma-ray and proton respectively, and subsequently the CBMN slides were cultured according to the preparation protocol. The results of automated and visual scoring for these CBMN samples were also compared finally.[Results]1. We standardized the slide preparation protocol in order to obtain an optimal density and dispersion of the cells, avoiding overlapping of the cells, which is of major importance for image analysis to reach the most appropriate detection as possible.2. All the key techniques are realized by means of image analysis and pattern recognition: (1) The quality of input CBMN images can be improved by the median filtering and morphological filtering; (2) The cell regions, nucleus regions and MN regions in the input CBMN images can be segmented by the iterative thresholding method; (3) Target regions in CBMN images can be classified by the characteristic parameters (Area, Elongation and Defect-ratio); (4) Clusters in the CBMN images can be separated efficiently and accurately by the improved watershed methods; (5) BNC and MN can be automated scored by means of analyzing the information within the boundaries of cells, i.e. the number of targets, the size and shape relationship of these targets, etc.3. An applied automated analysis system of CBMN images has been developed successfully based on the platform of MATLAB7.1 and GUI.4. We accomplished the performance testing of our automated analysis system with satisfying results, i.e. the detecting ability of BNC and MN are 82.42% and 73.89% respectively, and the recognizing ability of BNC and MN are 85.70% and 85.28%. The total analysis time of a whole slide is less than 3 hours.5. A series of applied algorithms for image analysis and pattern recognize have been developed during the automatization of CBMN image, such as (1) the robust double threshold segmentation method for grey-level images; (2) the alternate ultimate erosion method with the 4-neighborhood structure element (4-SE) and 8-SE instead of a single structure element; (3) the novel algorithm for separating the clumps by the improved watershed algorithm, and so on.[Conclusion]In this thesis all the key techniques and difficulties for automated detection and counting techniques of BNC and MN have been investigated and solved firstly by means of image processing and pattern recognition. And then an applied automated analysis system of CBMN images has been developed successfully, which can meet the needs of health assessment for large numbers of radiation workers and radiation dose estimation for victims after radiation accidence. This investigation is of high military significance and prominent social benefit.The boundary information of isolated (or separated) cells can be used to restrict the region of two nuclei of a BNC, and the clusters in the CBMN images can be separated efficiently, therefore the scoring ability and recognition accuracy of BNC and MN in our system have been improved evidently than the external systems. The total analysis time of a whole slide (within 3 hours) is also acceptable. The success of automatization of CBMN images also provides a guarantee for devising the automated image system for other kind of MN, as well as other medical image processing.In the future research, we will further optimize and perfect this automated analysis system of CBMN image to make it generally applicable.