developing an automated cytogenetic imaging system for detection of dicentric chromosomes in biological dosimetry

background: high-energy ionizing radiation is harmful and changes the genetic makeup of dna, which can lead to increased risk of cancer. thus, the exposure of radiation dose should be under control and limited. ionizing radiation might lead to some chromosome aberrations like dicentric. there is a strong relation between the frequency of dicentric chromosome in metaphase, and the received dose. objective: identifying the frequency of dicentric chromosomes is an important task especially in large-scale radiation accident for making rapid clinical decision. given this, a large number of metaphases should be investigated for an accurate estimating of dose. it is well known that non-automated (visual) scoring of chromosome aberrations such as dicentrics requires highly trained experts. on the other hand, as thousands of cells should be scored, it is time-consuming and results in fatigue that can lead to poor concentration. methods: biological dosimetry technique for assessing radiation dose is based on analyzing chromosome in peripheral blood lymphocyte. our study describes a technique to speed up this procedure by automatically distinguishing abnormal chromosomes from normal ones. the most important feature of dicentric chromosomes is their two centromeres.  therefore, the main approach in our study is to design an automated system to identify the number and position of centromeres. we presented a method for classifying chromosomes into four groups in accordance with the number of centromeres in each chromosome. results: the image dataset of 311 chromosomes of normal and dicentric chromosomes is used to test the scheme. the sensitivity about 90% and specificity more than 95% for classification of chromosomes into dicentric and non-dicentric as well as accuracy more than 90% for centromere identification, are obtained.  conclusion: if an automated system with the capability of straightening bent chromosome was applied prior to the proposed algorithm, even more accurate results would be achieved.