改进分水岭与主动轮廓模型相结合的肝癌细胞核分割算法研究

Nuclei Segmentation Combining Watershed with Active Contour Model in Liver Cancer Histopathology Images

肝癌已经成为常见的恶性肿瘤,组织切片显微图像的病理分析是诊断的主要手段,细胞的准确分割是病理分析的重要环节。提出了一种新的基于H&E染色组织病理学图像的肝癌细胞分割算法,首先用彩色卷积方法进行主成分提取得到灰度图像,在此灰度图像的基础上,进行形态学处理;然后运用快速径向对称变换提取种子点作为前景标记,进行分水岭分割;最后将分水岭分割得到的轮廓线作为初始轮廓进行GVF主动轮廓分割,得到更为精确的轮廓线。分割结果的阳性预测平均值达到0.89,召回率平均值达到0.9,证明了该算法细胞定位的准确性。Dice系数平均值为0.84,证明了本算法细胞分割区域的准确性。

Liver cancer has become a common malignant tumor in the world.Analysis of histopathology images is one of the main methods for diagnosis and precise segmentation of cell and nuclei is the first step towards automatic analysis of histopathology images.A new nuclei segmentation method that works with HE stained liver cancer histopathology images was proposed.The procedure can be concluded as the following steps：getting the grayscale image by color convolution,selecting seed points utilizing fast radial symmetry transform,marker-controlled watershed segmentation,getting final contour via active contour model.The evaluation was done in terms of positive predictive value,recall and Dice coefficient.The mean predictive value was 0.89 and the mean recall was 0.9 which indicates the accuracy of cell detection.The mean Dice coefficient was 0.84 which indicates the accuracy of segmentation.