1. School of Biomedical Engineering (Suzhou), University of Science and Technology of China, Suzhou 215163, China 2. Suzhou Institute of Biomedical Engineering and Technology, Chinese Academy of Sciences, Suzhou 215163, China 3. The Second Affiliated Hospital of Suzhou University, Suzhou 215000, China
A new deep convolutional neural network model called SE-Mask-RCNN was proposed to automatically segment prostate cancer regions in multi-parametric magnetic resonance imaging (mp-MRI) images. The candidate regions were extracted from the feature maps, and the possible lesions were segmented within the candidate regions. The feature maps were obtained from apparent diffusion coefficient (ADC) maps and T2-weighted (T2W) images through two parallel convolution networks, which were fused to fully use complementary information. The squeeze-and-excitation block was employed to automatically boost effective features and suppress the useless features. The experiments were conducted on a dataset containing 140 patients. The experimental results showed that the proposed model achieved a Dice coefficient of 0.654, a sensitivity of 0.695, a specificity of 0.970, and a positive predictive value of 0.685. The proposed SE-Mask-RCNN can effectively improve the segmentation accuracy of prostate cancer lesions in mp-MRI images compared with U-net, V-net, Resnet50-U-net and Mask-RCNN.
Fig.4mp-MRI information fusion method improved by SE-block
Fig.5Example of data preprocessing
网络模型
MRI类型
DSC
敏感度
特异度
PPV
U-net
ADC
0.539
0.639
0.970
0.324
V-net
ADC
0.525
0.707
0.945
0.472
Resnet50-U-net
ADC
0.525
0.720
0.956
0.462
Mask-RCNN
ADC
0.592
0.582
0.979
0.680
U-net
T2W
0.370
0.651
0.886
0.283
V-net
T2W
0.343
0.580
0.896
0.282
Resnet50-U-net
T2W
0.349
0.624
0.902
0.276
Mask-RCNN
T2W
0.403
0.395
0.962
0.473
U-net
ADC+T2W
0.562
0.765
0.955
0.480
V-net
ADC+T2W
0.553
0.763
0.949
0.476
Resnet50-U-net
ADC+T2W
0.561
0.683
0.972
0.532
Mask-RCNN
ADC+T2W
0.612
0.653
0.968
0.650
Tab.1Quantitative comparison of different network model results of prostate cancer lesion segmentation
Fig.6Examples of prostate cancer lesion segmentation results from different network models on mp-MRI
Fig.7Four methods of mp-MRI information fusion
融合方式
卷积网络
DSC
敏感度
特异度
PPV
输入图像串联(Mask-RCNN)
Resnet50
0.612
0.653
0.968
0.650
特征图相加
Resnet50
0.611
0.667
0.966
0.643
特征图串联
Resnet50
0.609
0.645
0.968
0.659
SE-特征图串联
Resnet50
0.630
0.676
0.970
0.663
输入图像串联
SE-Resnet
0.637
0.660
0.970
0.685
特征图相加
SE-Resnet
0.638
0.684
0.970
0.658
特征图串联
SE-Resnet
0.640
0.697
0.967
0.652
SE-特征图串联(SE-Mask-RCNN)
SE-Resnet
0.654
0.695
0.970
0.685
Tab.2Comparison of four methods of mp-MRI information fusion
Fig.8Comparison of candidate regions and heat maps
融合方式
ADC SE-Resnet是否去掉SE-block
T2W SE-Resnet是否去掉SE-block
DSC
敏感度
特异度
PPV
SE-特征图串联
否
否
0.654
0.695
0.970
0.685
SE-特征图串联
否
是
0.633
0.722
0.960
0.634
SE-特征图串联
是
否
0.633
0.726
0.960
0.630
SE-特征图串联
是
是
0.621
0.641
0.975
0.670
Tab.3Performance comparison of SE-block removal methods for convolutional networks
模型
运算时间/ms
U-net
78
V-net
84
Resnet50-U-net
85
Mask-RCNN
131
SE-Mask-RCNN
144
Tab.4Average image processing time for different models
[1]
CHEN W, ZHENG R, BAADE P D, et al Cancer statistics in China, 2015[J]. CA: A Cancer Journal for Clinicians, 2016, 66 (2): 115- 132
doi: 10.3322/caac.21338
[2]
CHEN W, ZHENG R, ZENG H, et al Annual report on status of cancer in China, 2011[J]. Chinese Journal of Cancer Research, 2015, 27 (1): 2
[3]
陈金东 中国各类癌症的发病率和死亡率现状及发展趋势[J]. 遵义医学院学报, 2018, 41 (6): 653- 662 CHEN Jin-dong Trends of cancer incidence and mortality in China[J]. Journal of Zunyi Medical University, 2018, 41 (6): 653- 662
doi: 10.3969/j.issn.1000-2715.2018.06.001
[4]
王宁, 刘硕, 杨雷, 等 2018全球癌症统计报告解读[J]. 肿瘤综合治疗电子杂志, 2019, 5 (1): 87- 97 WANG Ning, LIU Shuo, YANG Lei, et al Interpretation on the report of global cancer statistics 2018[J]. Journal of Multidisciplinary Cancer Management (Electronic Version), 2019, 5 (1): 87- 97
[5]
SIEGEL R L, MILLER K D, JEMAL A Cancer statistics, 2019[J]. CA: A Cancer Journal for Clinicians, 2019, 69 (1): 7- 34
doi: 10.3322/caac.21551
[6]
PANEBIANCO V, BARCHETTI F, SCIARRA A, et al Multiparametric magnetic resonance imaging vs. standard care in men being evaluated for prostate cancer: a randomized study[J]. Urologic Oncology: Seminars and Original Investigations, 2015, 33 (1): 17.e1- 17.e7
doi: 10.1016/j.urolonc.2014.09.013
[7]
FüTTERER J J, BRIGANTI A, DE VISSCHERE P, et al Can clinically significant prostate cancer be detected with multiparametric magnetic resonance imaging? a systematic review of the literature[J]. European Urology, 2015, 68 (6): 1045- 1053
doi: 10.1016/j.eururo.2015.01.013
[8]
王建业, 陈鑫, 刘明, 等 MRI在前列腺癌诊断和治疗中的应用[J]. 磁共振成像, 2010, 1 (4): 253- 256 WANG Jian-ye, CHEN Xin, LIU Ming, et al MRI in clinical diagnosis and treatment of prostate cancer[J]. Chinese Journal of Magnetic Resonance Imaging, 2010, 1 (4): 253- 256
[9]
KOHL S, BONEKAMP D, SCHLEMMER H P, et al. Adversarial networks for the detection of aggressive prostate cancer [DB/OL]. [2017-02-26]. https://arxiv.org/abs/1702.08014.
[10]
蒋宏达, 叶西宁 一种改进的I-Unet网络的皮肤病图像分割算法[J]. 现代电子技术, 2019, 42 (12): 52- 56 JIANG Hong-da, YE Xi-ning An improved skin disease image segmentation algorithm based on I-Unet network[J]. Modern Electronics Technique, 2019, 42 (12): 52- 56
邢波涛, 李锵, 关欣 改进的全卷积神经网络的脑肿瘤图像分割[J]. 信号处理, 2018, 34 (8): 911- 922 XING Bo-tao, LI Qiang, GUAN Xin A brain tumor image segmentation method based on improved fully convolutional neural network[J]. Journal of Signal Processing, 2018, 34 (8): 911- 922
[13]
田萱, 王亮, 丁琪 基于深度学习的图像语义分割方法综述[J]. 软件学报, 2019, 30 (2): 250- 278 TIAN Xuan, WANG Liang, DING Qi Review of image semantic segmentation based on deep learning[J]. Journal of Software, 2019, 30 (2): 250- 278
[14]
RONNEBERGER O, FISCHER P, BROX T. U-net: convolutional networks for biomedical image segmentation [C] // International Conference on Medical Image Computing and Computer-Assisted Intervention. Munich: Springer, 2015: 234-241.
[15]
HE K, GKIOXARI G, DOLLáR P, et al. Mask R-CNN [C] // IEEE International Conference on Computer Vision. Washington D. C. : IEEE, 2017: 2961-2969.
[16]
TSEHAY Y K, LAY N S, ROTH H R, et al. Convolutional neural network based deep-learning architecture for prostate cancer detection on multi-parametric magnetic resonance images [C] // Medical Imaging 2017: Computer-Aided Diagnosis. Orlando: SPIE, 2017: 1013405.
[17]
YANG X, LIU C, WANG Z, et al Co-trained convolutional neural networks for automated detection of prostate cancer in multi-parametric MRI[J]. Medical Image Analysis, 2017, 42: 212- 227
doi: 10.1016/j.media.2017.08.006
[18]
YUAN Y, QIN W, BUYYOUNOUSKI M, et al Prostate cancer classification with multiparametric MRI transfer learning model[J]. Medical Physics, 2019, 46 (2): 756- 765
doi: 10.1002/mp.13367
[19]
HU J, SHEN L, SUN G. Squeeze-and-excitation networks [C] // Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City: IEEE, 2018: 7132-7141.
[20]
LIU P, WANG S, TURKBEY B, et al. A prostate cancer computer-aided diagnosis system using multimodal magnetic resonance imaging and targeted biopsy labels [C] // Medical Imaging 2013: Computer-Aided Diagnosis. Lake Buena Vista: SPIE, 2013: 86701G.
[21]
YU Q, LIU F, SONG Y Z, et al. Sketch me that shoe [C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 799-807.
[22]
QIAN C, WANG L, GAO Y, et al In vivo MRI based prostate cancer localization with random forests and auto-context model[J]. Computerized Medical Imaging and Graphics, 2016, 52: 44- 57
doi: 10.1016/j.compmedimag.2016.02.001
[23]
MAZZETTI S, GIANNINI V, RUSSO F, et al Computer-aided diagnosis of prostate cancer using multi-parametric MRI: comparison between PUN and Tofts models[J]. Physics in Medicine and Biology, 2018, 63 (9): 095004
doi: 10.1088/1361-6560/aab956
[24]
MILLETARI F, NAVAB N, AHMADI S A. V-net: fully convolutional neural networks for volumetric medical image segmentation [C]// 2016 4th International Conference on 3D Vision. Stanford: IEEE, 2016: 565-571.
[25]
ISHIOKA J, MATSUOKA Y, UEHARA S, et al Computer‐aided diagnosis of prostate cancer on magnetic resonance imaging using a convolutional neural network algorithm[J]. BJU International, 2018, 122 (3): 411- 417
doi: 10.1111/bju.14397
[26]
陆洋, 柏根基 定量动态增强磁共振成像及其在前列腺癌诊断和治疗中的研究进展[J]. 磁共振成像, 2015, 6 (10): 796- 800 LU Yang, BO Gen-ji Progress of quantitative dynamic contrast-enhanced magnetic resonance imaging and in diagnosis and evaluation on the therapeutic response of prostate cancer[J]. Chinese Journal of Magnetic Resonance Imaging, 2015, 6 (10): 796- 800
doi: 10.3969/j.issn.1674-8034.2015.10.016