基于深度卷积神经网络的肾透明细胞癌细胞核分割

发布时间：2018-03-18 03:07

本文选题：分割　切入点：卷积神经网络　出处：《生物医学工程研究》2017年04期 　论文类型：期刊论文

【摘要】：肾透明细胞癌病理图像中细胞核的形态和位置信息对肾癌的良恶性分级诊断具有重要意义,为提高肾透明细胞癌细胞核分割的质量,本研究提出了基于深度卷积神经网络的细胞核分割方法。首先,根据标定的病理图像中细胞核轮廓,构建细胞核分割样本集;然后,深度卷积神经网络通过隐式特征学习对细胞核分割模型进行训练,避免人为设计特征;最后,利用细胞核分割模型对病理图像进行逐像素分割。实验结果表明,深度卷积神经网络的细胞核分割算法在肾透明细胞癌细胞核分割的像素准确率高达90.33%,细胞核分割性能稳定,深度卷积神经网络强大的鲁棒性和适应性使得肾透明细胞癌细胞核自动分割具有可能。
[Abstract]:The morphological and location information of the nuclei in the pathological images of renal clear cell carcinoma is of great significance for the diagnosis of benign and malignant grade of renal cell carcinoma, in order to improve the quality of nuclear segmentation in renal clear cell carcinoma. In this study, a nuclear segmentation method based on deep convolution neural network is proposed. Firstly, according to the nuclear contour in the calibrated pathological image, the nuclear segmentation sample set is constructed. The deep convolution neural network trains the nuclear segmentation model by implicit feature learning to avoid artificial design features. Finally, we use the nuclear segmentation model to segment the pathological image pixel by pixel. The experimental results show that, The pixel segmentation accuracy of deep convolution neural network in renal clear cell carcinoma is as high as 90.33, and the cell nuclear segmentation performance is stable. The strong robustness and adaptability of deep convolution neural networks make it possible to segment the nuclei automatically in renal clear cell carcinoma.
【作者单位】：南京信息工程大学江苏省大数据分析技术重点实验室;南京军区南京总医院;
【基金】：国家自然科学基金资助项目(61273259) 江苏省自然科学基金资助项目(BK20141482)
【分类号】：R737.11;TP183

【相似文献】