Rapid and Automatic Detection of Micronuclei in Binucleated lymphocytes Image

In terms of radiation biological dose estimation, the cytokinesis block micronucleus (CBMN) assay is the internationally recognized dose estimation method. Due to the subjectivity and the time-consuming of manual detection, it cannot meet the needs of rapid standard assay of CBMN. Therefore, in this research work, we combined the convolutional neural network to design a software that can be used for rapid standard automatic detection of micronuclei in Giemsa stained binucleated lymphocytes image. The software analysis workflow is divided into four stages: cell acquisition, adhesive cell masses segmentation, cell type identification, micronucleus counting. After verification, our algorithm can quickly and effectively detect binucleated cells and micronucleus even when the cytoplasm is blurred, multiple micronucleus are attached to each other, or micronucleus is attached to the nucleus. In the test of a large number of random images, the software reached 99.4% of the manual detection in terms of the detection rate of binucleated cell, and the false positive rate of binucleated cell was 14.7%. In terms of micronucleus detection rate, the software reached 115.1% of manual detection, and its false positive rate was 26.2%. The analysis time of each picture is about 0.3s, an order of magnitude faster than conventional method.