新型乳腺磁共振增强图像肿瘤区域的自动分割模型
|
摘要
乳腺磁共振增强图像上,乳腺癌主要有肿块型和非肿块型两种强化方式。由于乳腺肿瘤区域相对较小,肿块型和非肿块型之间形态学差异大,非肿块型自身差异性复杂,因而很难精确分割出乳腺肿瘤区域。针对这些问题,提出一套新颖的粗检测细分割的深度学习模型(YOLOv2+SegNet)。该模型在精准分割之前,首先运用YOLOv2网络在乳腺可能的肿瘤区域进行粗检测,从而得到大致可能的肿瘤区域;接下来在粗检测的基础上,针对检测到可能的肿瘤区域,运用SegNet网络进行精细分割,从而实现算法最优的性能。为了验证YOLOv2+SegNet模型的有效性,从医院采集的数据集中选取560张乳腺MRI增强图像作为训练和测试(其中训练和测试集分别为415张和145张乳腺MRI数据)。在实验的过程中,运用YOLOv2+SegNet模型,分别对乳腺肿块型、非肿块型、肿块和非肿块混合型3类MRI数据进行肿瘤区域自动分割的实验。实验结果表明:YOLOv2+SegNet模型和SegNet网络分割结果的Dice系数相比有约10%的提升,与传统的C-V模型、模糊C均值聚类、光谱映射主动轮廓模型以及深度模型U-net、DeepLab相比有更为明显的提升。
Breast cancer can be mainly classified into two kinds: mass-like and non-mass-like on enhanced breast images. Owing to the small area of breast cancer, along with the huge difference between the shape of mass-like and non-mass-like and the self complexity of non-mass-like, it is hard to segment the accurate area of breast tumor. To solve these problems, this paper proposed a novel deep learning model of rough detection and fine segmentation. Before precise segmentation, rough detection for the cancer region was firstly processed for potential region of the tumor. On the basis of rough detection, we used SegNet for fine segmentation to achieve the best performance of the algorithm. In order to test the effectiveness of proposed method(YOLOv2+SegNet), we picked 560 magnetic resonance imaging(MRI) images of breast cance out of the dataset collected from the hospital for training and testing(415 images for training and 145 for testing). For more comprehensive analysis, experiments were set to analyze three different conditions, such as mass-like, non-mass-like and the mix of mass-like and non-mass-like. From the results, the established method improved 10% under each condition and improved a lot compared with the traditional C-V model, fuzzy C mean clustering, active contour model for spectral mapping and deep model of U-net or DeepLab.
Breast cancer can be mainly classified into two kinds: mass-like and non-mass-like on enhanced breast images. Owing to the small area of breast cancer, along with the huge difference between the shape of mass-like and non-mass-like and the self complexity of non-mass-like, it is hard to segment the accurate area of breast tumor. To solve these problems, this paper proposed a novel deep learning model of rough detection and fine segmentation. Before precise segmentation, rough detection for the cancer region was firstly processed for potential region of the tumor. On the basis of rough detection, we used SegNet for fine segmentation to achieve the best performance of the algorithm. In order to test the effectiveness of proposed method(YOLOv2+SegNet), we picked 560 magnetic resonance imaging(MRI) images of breast cance out of the dataset collected from the hospital for training and testing(415 images for training and 145 for testing). For more comprehensive analysis, experiments were set to analyze three different conditions, such as mass-like, non-mass-like and the mix of mass-like and non-mass-like. From the results, the established method improved 10% under each condition and improved a lot compared with the traditional C-V model, fuzzy C mean clustering, active contour model for spectral mapping and deep model of U-net or DeepLab.
引文
[1] Siegel RL, Miller KD, Jemal A. Cancer statistics, 2016[J]. CA: A Cancer Journal for Clinicians, 2016, 66(1): 7-30.
[2] Chen Wanqing, Zheng Rongshou, Baade PD, et al. Cancer statistics in China, 2015[J]. CA: A Cancer Journal for Clinicians, 2016, 66(2): 115-132.
[3] Moftah HM, Azar AT, Al-Shammari ET, et al. Adaptive k-means clustering algorithm for MR breast image segmentation[J]. Neural Computing & Applications, 2014, 24(7-8):1917-1928.
[4] Aslam A, Khan E, Beg MMS. Improved edge detection algorithm for brain tumor segmentation[J]. Procedia Computer Science, 2015, 58:430-437.
[5] Alfaris AQ, Ngah UK, Isa NAM, et al. Computer-aided segmentation system for breast MRI tumour using modified automatic seeded region growing (BMRI-MASRG)[J]. Journal of Digital Imaging, 2014, 27(1):133.
[6] Zadeh HG, Haddadnia J, Montazeri A. A model for diagnosing breast cancerous tissue from thermal images using active contour and lyapunov exponent[J]. Iranian Journal of Public Health, 2016, 45(5):657-669.
[7] Li Bang Nan, Chui Chee Kong, et al. Integrating spatial fuzzy clustering with level set methods for automated medical image segmentation[J]. Computers in Biology and Medicine, 2011, 41(1): 1-10.
[8] Tuwohingide D, Fatichah C. Spatial Fuzzy C-means dan rapid region merging untuk pemisahan sel kanker payudara[J]. Jurnal Nasional Teknik Elektro dan Teknologi Informasi (JNTETI), 2017, 6(1):294-294.
[9] Su Hai, Liu Fujun, Xie Yuanpu, et al. Region segmentation in histopathological breast cancer images using deep convolutional neural network[C]//International Symposium on Biomedical Imaging. California:IEEE, 2015:55-58.
[10] Wang J, Zhang Z, Li B, et al. An enhanced fall detection system for elderly person monitoring using consumer home networks[J]. IEEE Transactions on Consumer Electronics, 2014, 60(1): 23-29.
[11] He Kaiming, Gkioxari G, Dollar P, et al. Mask R-CNN[J]. arXiv:1703.06870,2017.
[12] MacMahon H, Naidich DP, Goo JM, et al. Guidelines for management of incidental pulmonary nodules detected on CT images: From the Fleischner Society 2017[J]. Radiology, 2017, 284(1): 228-243.
[13] Redmon J, Farhadi A. YOLO9000: Better, faster, stronger[EB/OL]. https://arxiv.org/cs/1612.08242. 2016-12-05/2017-10-04.
[14] Redmon J, Divvala S, Girshick R, et al. You only look once: unified, real-time object detection[C]// Computer Vision and Pattern Recognition. California: IEEE, 2016:779-788.
[15] Noh H, Hong S, Han B. Learning deconvolution network for semantic segmentation[C]//Proceedings of the IEEE International Conference on Computer Vision. California:IEEE, 2015: 1520-1528.
[16] Ren Shaoqing, Girshick R, Girshick R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, 39(6):1137-1149.
[17] Hartigan JA, Wong MA. Algorithm AS 136: A K-means clustering algorithm[J]. Journal of the Royal Statistical Society, 1979, 28(1):100-108.
[18] Kass M, Witkin A, Terzopoulos D. Snakes: Active contour models[J]. International Journal of Computer Vision, 1988, 1(4):321-331.
[19] Xu Jun, Janowczyk A, Chandran S, et al. A high-throughput active contour scheme for segmentation of histopathological imagery[J]. Medical Image Analysis, 2011, 15(6):851-862.
[20] Badrinarayanan V, Kendall A, Cipolla R. SegNet: A deep convolutional encoder-decoder architecture for scene segmentation[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, 39:2481-2495.
[21] Salamon J, Bello JP. Deep convolutional neural networks and data augmentation for environmental sound classification[J]. IEEE Signal Processing Letters, 2017, 24: 279-283.
[22] CireAn D, Meier U, Masci J, et al. Multi-column deep neural network for traffic sign classification[J]. Neural Networks, 2012, 32: 333-338.
[23] Fedorov A, Beichel R, Kalpathycramer J, et al. 3D Slicer as an image computing platform for the quantitative imaging network[J]. Magnetic Resonance Imaging, 2012, 30:1323-1341.
[24] Boegel M, Hoelter P, Redel T, et al. A fully-automatic locally adaptive thresholding algorithm for blood vessel segmentation in 3D digital subtraction angiography[C]//?The 37th Annual International Conference of the IEEE EMBS. Milan: IEEE, 2015:2006-2009.
[25] S?rlie T, Perou CM, Tibshirani R, et al. Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications[J]. PNAS, 2001, 98:10869-10874.
[26] S?rlie T, Tibshirani R, Parker J, et al. Repeated observation of breast tumor subtypes in independent gene expression data sets[J]. PNAS, 2003, 100:8418-8423.
[2] Chen Wanqing, Zheng Rongshou, Baade PD, et al. Cancer statistics in China, 2015[J]. CA: A Cancer Journal for Clinicians, 2016, 66(2): 115-132.
[3] Moftah HM, Azar AT, Al-Shammari ET, et al. Adaptive k-means clustering algorithm for MR breast image segmentation[J]. Neural Computing & Applications, 2014, 24(7-8):1917-1928.
[4] Aslam A, Khan E, Beg MMS. Improved edge detection algorithm for brain tumor segmentation[J]. Procedia Computer Science, 2015, 58:430-437.
[5] Alfaris AQ, Ngah UK, Isa NAM, et al. Computer-aided segmentation system for breast MRI tumour using modified automatic seeded region growing (BMRI-MASRG)[J]. Journal of Digital Imaging, 2014, 27(1):133.
[6] Zadeh HG, Haddadnia J, Montazeri A. A model for diagnosing breast cancerous tissue from thermal images using active contour and lyapunov exponent[J]. Iranian Journal of Public Health, 2016, 45(5):657-669.
[7] Li Bang Nan, Chui Chee Kong, et al. Integrating spatial fuzzy clustering with level set methods for automated medical image segmentation[J]. Computers in Biology and Medicine, 2011, 41(1): 1-10.
[8] Tuwohingide D, Fatichah C. Spatial Fuzzy C-means dan rapid region merging untuk pemisahan sel kanker payudara[J]. Jurnal Nasional Teknik Elektro dan Teknologi Informasi (JNTETI), 2017, 6(1):294-294.
[9] Su Hai, Liu Fujun, Xie Yuanpu, et al. Region segmentation in histopathological breast cancer images using deep convolutional neural network[C]//International Symposium on Biomedical Imaging. California:IEEE, 2015:55-58.
[10] Wang J, Zhang Z, Li B, et al. An enhanced fall detection system for elderly person monitoring using consumer home networks[J]. IEEE Transactions on Consumer Electronics, 2014, 60(1): 23-29.
[11] He Kaiming, Gkioxari G, Dollar P, et al. Mask R-CNN[J]. arXiv:1703.06870,2017.
[12] MacMahon H, Naidich DP, Goo JM, et al. Guidelines for management of incidental pulmonary nodules detected on CT images: From the Fleischner Society 2017[J]. Radiology, 2017, 284(1): 228-243.
[13] Redmon J, Farhadi A. YOLO9000: Better, faster, stronger[EB/OL]. https://arxiv.org/cs/1612.08242. 2016-12-05/2017-10-04.
[14] Redmon J, Divvala S, Girshick R, et al. You only look once: unified, real-time object detection[C]// Computer Vision and Pattern Recognition. California: IEEE, 2016:779-788.
[15] Noh H, Hong S, Han B. Learning deconvolution network for semantic segmentation[C]//Proceedings of the IEEE International Conference on Computer Vision. California:IEEE, 2015: 1520-1528.
[16] Ren Shaoqing, Girshick R, Girshick R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, 39(6):1137-1149.
[17] Hartigan JA, Wong MA. Algorithm AS 136: A K-means clustering algorithm[J]. Journal of the Royal Statistical Society, 1979, 28(1):100-108.
[18] Kass M, Witkin A, Terzopoulos D. Snakes: Active contour models[J]. International Journal of Computer Vision, 1988, 1(4):321-331.
[19] Xu Jun, Janowczyk A, Chandran S, et al. A high-throughput active contour scheme for segmentation of histopathological imagery[J]. Medical Image Analysis, 2011, 15(6):851-862.
[20] Badrinarayanan V, Kendall A, Cipolla R. SegNet: A deep convolutional encoder-decoder architecture for scene segmentation[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, 39:2481-2495.
[21] Salamon J, Bello JP. Deep convolutional neural networks and data augmentation for environmental sound classification[J]. IEEE Signal Processing Letters, 2017, 24: 279-283.
[22] CireAn D, Meier U, Masci J, et al. Multi-column deep neural network for traffic sign classification[J]. Neural Networks, 2012, 32: 333-338.
[23] Fedorov A, Beichel R, Kalpathycramer J, et al. 3D Slicer as an image computing platform for the quantitative imaging network[J]. Magnetic Resonance Imaging, 2012, 30:1323-1341.
[24] Boegel M, Hoelter P, Redel T, et al. A fully-automatic locally adaptive thresholding algorithm for blood vessel segmentation in 3D digital subtraction angiography[C]//?The 37th Annual International Conference of the IEEE EMBS. Milan: IEEE, 2015:2006-2009.
[25] S?rlie T, Perou CM, Tibshirani R, et al. Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications[J]. PNAS, 2001, 98:10869-10874.
[26] S?rlie T, Tibshirani R, Parker J, et al. Repeated observation of breast tumor subtypes in independent gene expression data sets[J]. PNAS, 2003, 100:8418-8423.