Research On Automatic Human Chromosome Image Analysis

Human chromosome karyotyping is one of the essential tasks in cytogenetics, especially in genetic syndrome diagnoses. It refers to the classification and subsequently a formatted display of the chromosomes found in a cell spread. The traditional chromosome karyotyping is performed by hand, which is time consuming and exhausting. In this thesis, an automatic procedure is introduced for human chromosome image analysis. The important technologies and core algorithms used in the procedure are discussed in detail as follows:(1) Automatic segmentation of touching and overlapping chromosomes is one of the most important parts. According to different status of touching and overlapping chromosomes, different methods are proposed. The inner holes are eliminated by the dividing lines linked by the two nearest concave points of the inner and external borders. The watershed algorithm is used to segment narrow-touching chromosomes. Another algorithm to detect and segment touching and overlapping chromosomes based on geometric features is presented, and the pale-path is searched for touching chromosomes segmentation.(2) Feature extraction is also one of the most important parts. Medial axis is extracted by the middle point algorithm, and is optimized according to different cases. Centromere is located based on geometric features and gray distribution. Chromosome band is enhanced by the algorithm based on multiscale B-spline wavelets, extracted by average gray profile, gradient profile and shape profile, and calculated by the WDD (Weighted Density Distribution) descriptors.(3) Automatic chromosome recognition and classification is another most important part. The multilayer classifier is used. In the first layer, the chromosomes are classified into several groups by geometric classification method. In the second layer, chromosomes in each group are classified by Bayesian classifier.The experiment results demonstrate that the algorithms perform well.