Aiming at the problem of low efficiency of dicentric chromosome identification counting under the microscope, this paper presents a joint processing algorithm combining clustering and watershed. The method first uses clustering and watershed algorithm to segment the original chromosome image, and then identifies the individual chromosomes. The results show that when the equivalent width Y parameter is selected m = 1, n = 1, the true positive rate of dicentric chromosome identification is 76.6%, and positive predictive value is 76.6% in high dose, which is higher than the threshold algorithm for the true positive rate (63.9%) and positive predictive value (63.5%). The number of identified dicentric chromosomes can be used for dose estimation. When 500 cells are used for identification and dose estimation, the dose estimation pass rate can reach 80% in high dose. But for low dose, more cells should be used to identify to increase the dose estimation pass rate.

针对显微镜下双中心染色体识别计数效率低的问题，提出了一种结合聚类和分水岭的联合处理算法。该方法首先使用聚类和分水岭算法对原始染色体图像进行分割，然后识别单个染色体。结果表明，当等效宽度Y参数选择M = 1，n = 1时，的双着丝粒染色体鉴定真阳性率是76.6％，阳性预测值76.6％，高剂量，这是比阈值算法更高真实阳性率（63.9％）和阳性预测值（63.5％）。所识别的双着丝粒染色体的数量可以用于剂量估计。当使用500个细胞进行鉴定和剂量估算时，高剂量时剂量估算合格率可达到80％。但是对于低剂量，应该使用更多的细胞来识别，以增加剂量估计的通过率。