基于弱监督宫颈细胞图像的语义分割方法

doi:10.16180/j.cnki.issn1007-7820.2021.12.012

摘要/Abstract

摘要：

基于弱监督标签的神经网络算法是医学领域的研究热点之一。针对医学领域缺少标签数据以及细胞质、细胞核分割不精确的问题,文中提出一种基于弱监督宫颈细胞图像的语义分割算法。该算法首先用无监督K-means作为标注函数生成细胞图像分割标签。然后,通过改进的Encoder-Decoder网络进行训练。文中引入CRF作为网络的最后一层以整合图片全局信息,优化分割结果。将标签及预测图像分3次进行优化训练以达到对细胞图像的像素级分类。在宫颈细胞图片数据集上对文中所提算法进行验证,实验结果表明,该算法具有良好的泛化能力,准确率高达96.7%。

关键词: 弱监督, K-means, 语义分割, 卷积神经网络, 宫颈细胞, Encoder-Decoder, CRF, 优化训练

Abstract:

Neural network algorithm based on weakly supervised label is a hot research topic in medical field. In view of the lack of labeling data and the inaccurate segmentation of cytoplasm and nucleus, this study proposes a semantic segmentation algorithm based on weakly supervised cervical cell images. First, the algorithm uses unsupervised K-means as the labeling function to generate cell image segmentation labels. Then, training is conducted through an improved Encoder-Decoder network. Subsequently, CRF is introduced as the last layer of the network to integrate the global information of the image and optimize the segmentation results. The label and prediction images are optimized and trained in three times to achieve pixel level classification of cell images. Finally, the algorithm is verified using the cervical cell image dataset. The experimental results show that the algorithm has high generalization ability, and the accuracy rate is up to 96.7%.

Key words: weak supervision, K-means, semantic segmentation, convolutional neural network, cervical cell, Encoder-Decoder, CRF, optimization training

中图分类号:

TP391

张灿,陈玮,尹钟. 基于弱监督宫颈细胞图像的语义分割方法[J]. 电子科技, 2021, 34(12): 68-74.

ZHANG Can,CHEN Wei,YIN Zhong. Semantic Segmentation of Cervical Cell Image Based on Weak Supervision[J]. Electronic Science and Technology, 2021, 34(12): 68-74.

图/表 8

图1

图2

图3

图4

图5

图6

表1

表2

参考文献 25

[1]	杨志明, 李亚伟, 杨冰, 等. 融合宫颈细胞领域特征的多流卷积神经网络分类算法[J]. 计算机辅助设计与图形学学报, 2019, 31(4):531-540.
	Yang Zhiming, Li Yawei, Yang Bing, et al. Multi flow convolution neural network classification algorithm based on the characteristics of cervical cell field[J]. Journal of Computer-Aided Design & Computer Graphics, 2019, 31(4):531-540.
[2]	于月娜. 宫颈细胞图像分割和特征提取算法研究[D]. 湘潭:湘潭大学, 2019.
	Yu Yuena. Research on segmentation and feature extraction algorithms cervical cell images[D]. Xiangtan:Xiangtan University, 2019.
[3]	段鹏, 程文播, 钱庆, 等. 基于瓶颈检测和分水岭算法的重叠宫颈细胞图像分割方法[J]. 中国医疗器械杂志, 2020, 44(1):7-12.
	Duan Peng, Cheng Wenbo, Qian Qing, et al. Overlapping cervical cell Image segmentation based on bottleneck detection and watershed algorithm[J]. Chinese Journal of Medical Instrumentation, 2020, 44(1):7-12.
[4]	孟楚楚, 赵立宏. 结合梯度边缘信息改进的全局阈值法与GVF Snake模型的宫颈细胞图像分割[J]. 智能计算机与应用, 2019, 9(2):28-32.
	Meng Chuchu, Zhao Lihong. Segmentation of cervical cell image using improved global thresholding method with gradient edge information and GVF Snake model[J]. Intelligent Computer and Applications, 2019, 9(2):28-32.
[5]	Lassouaoui N, Hamami L. Genetic algorithms and multifractal segmentation of cervical cell images [C].Paris: The Seventh International Symposium on Signal Processing and Its Applications, 2003.
[6]	李智能, 刘任任, 梁光明. 基于卷积神经网络的医学宫颈细胞图像的语义分割[J]. 计算机应用与软件, 2019, 36(11):152-156.
	Li Zhineng, Liu Renren, Liang Guangming. Semantic segmentation of medical cervical cell images based on convolutional neural network[J]. Computer Applications and Software, 2019, 36(11):152-156.
[7]	Richmond D L, Kainmueller D, Yang M Y, et al. Relating cascaded random forests to deep convolutional neural networks for semantic segmentation[J]. Computer Science, 2015, 36(12):233-245.
[8]	Micha J, Rgen K, Li P H, et al. High-precision automated reconstruction of neurons with flood-filling networks[J]. Nature Methods, 2018, 15(18):605-610. doi: 10.1038/s41592-018-0049-4
[9]	Wei Y, Liang X, Chen Y, et al. Stc:a simple to complex framework for weakly-supervised semantic segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2016, 39(11):2314-2320. doi: 10.1109/TPAMI.2016.2636150
[10]	Jiao L, Huo L, Hu C, et al. Refined UNet V2: end-to-end patch-wise network for noise-free cloud and shadow segmentation[J]. Remote Sensing, 2020, 12(21):3530-3566. doi: 10.3390/rs12213530
[11]	Qi C R, Su H, Mo K, et al. Pointnet:deep learning on point sets for 3D classification and segmentation [C].Honolulu:IEEE Conference on Computer Vision and Pattern Recognition, 2017.
[12]	Badrinarayanan V, Kendall A, Cipolla R. SegNet: a deep convolutional encoder-decoder architecture for image segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(12):2481-2495. doi: 10.1109/TPAMI.2016.2644615 pmid: 28060704
[13]	Zheng S, Jayasumana S, Romera-Paredes B, et al. Conditional random felds as recurrent neural network [C]. Santiago:Proceedings of the IEEE International Conference on Computer Vision, 2015.
[14]	Wang Y J, Hu S Y, Wang G D, et al. Multi-scale dilated convolution of convolutional neural network for crowd counting[J]. Multimedia Tools and Applications, 2020, 79(5):1057-1073. doi: 10.1007/s11042-019-08208-6
[15]	Pinheiro P O, Lin T Y, Collobert R, et al. Learning to refine object segments [C].Amsterdam:European Conference on Computer Vision, 2016.
[16]	Prasad R, Chang X J, Cloudhury S, et al. Multi-stage cascaded deconvolution for depth map and surface normal prediction from single image[J]. Pattern Recognition Letters, 2019, 27(16):165-173.
[17]	Ahron M, Elad M, Bruckstein A. K-SVD:an algorithm for designing overcomplete dictionaries for sparse representation[J]. IEEE Transactions on Signal Processingm, 2006, 54(11):4311-4322.
[18]	刘仪. 基于卷积神经网络的低分辨率细胞图像分割算法研究[D]. 西安:西安理工大学, 2019.
	Liu Yi. Research on low resolution cell images segmentation based on convolution neural networks[D]. Xi'an:Xi'an University of Technology, 2019.
[19]	Kothari S C, Prabhu G M, Roberts R. A multipath network with cross links[J]. Journal of Parallel & Distributed Computing, 1988, 5(2):185-193.
[20]	Shelhamer E, Long J, Darrell T. Fully convolutional networks for semantic segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 39(4):640-651. doi: 10.1109/TPAMI.2016.2572683
[21]	Badrinarayanan V, Kendall A, Cipolla R. Segnet:a deep convolutional encoder-decoder architecture for image segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(12):2481-2495. doi: 10.1109/TPAMI.2016.2644615 pmid: 28060704
[22]	Zhou Z W, Siddiquee M M R, Tajbakhsh N, et al. UNet++:redesigning skip connections to exploit multiscale features in image segmentation[J]. IEEE Transactions on Medical Imaging, 2019, 39(6):1856-1867. doi: 10.1109/TMI.42
[23]	Wei R, Liu B, Zhou F G. A patient-independent CT intensity matching method using conditional Generative Adversarial Networks (cGAN) for single X-ray projection based tumor localization[J]. Physics in Medicine & Biology, 2020,65(14):145009(1-16).
[24]	陆超文, 李菲菲, 陈虬. 基于改进哈希算法的图像检索方法[J]. 电子科技, 2020, 33(5):28-32.
	Lu Chaowen, Li Feifei, Chen Qiu. An image retrieval algorithm based on improved hashing method[J]. Electronic Science and Technology, 2020, 33(5):28-32.
[25]	严茂声, 陈家琪. 一种多环境适用的交通标志识别模型[J]. 电子科技, 2019, 32(4):71-76.
	Yan Maosheng, Chen Jiaqi. A multi-environmental traffic sign recognition model[J]. Electronic Science and Technology, 2019, 32(4):71-76.

模型	1	2	3	4	5	6	7	8	9	10	平均值
FCN	87.4%	88.9%	88.4%	80.8%	87.8%	66.8%	90.0%	77.5%	90.0%	91.6%	84.9%
SegNet	84.3%	89.6%	92.3%	92.9%	96.2%	88.7%	92.7%	94.0%	93.7%	88.9%	91.3%
UNet++	93.3%	74.8%	94.7%	96.0%	95.3%	69.3%	96.3%	95.7%	78.2%	92.2%	88.9%
CGAN	91.6%	93.0%	87.1%	87.8%	88.9%	74.8%	89.6%	86.6%	86.0%	88.9%	87.4%
EDCNet	97.6%	97.5%	97.1%	97.0%	98.1%	97.0%	97.6%	96.5%	98.1%	98.1%	97.5%

模型	1	2	3	4	5	6	7	8	9	10	平均值
FCN	80.9%	78.6%	82.4%	78.1%	80.3%	61.0%	81.9%	74.5%	80.9%	80.8%	77.9%
SegNet	80.5%	67.2%	83.8%	88.2%	91.7%	79.2%	87.4%	89.0%	85.2%	58.7%	81.1%
UNet++	76.2%	71.9%	92.7%	93.4%	93.4%	67.6%	93.6%	93.1%	75.9%	84.2%	84.2%
CGAN	85.9%	87.2%	85.0%	85.1%	86.7%	71.4%	85.0%	83.9%	82.8%	82.7%	83.6%
EDCNet	96.7%	96.7%	96.7%	96.2%	97.6%	96.2%	96.7%	95.7%	97.2%	96.9%	96.7%