基于多尺度特征与通道特征融合的脑肿瘤良恶性分类模型

姜林奇; 宁春玉; 余海涛

doi:10.37188/CO.2022-0067

基于多尺度特征与通道特征融合的脑肿瘤良恶性分类模型

doi: 10.37188/CO.2022-0067

长春理工大学生命科学技术学院, 吉林长春 130022

基金项目: 吉林省科技发展计划项目（No. 20200404219YY）

详细信息

作者简介:
姜林奇（1999—），女，内蒙古呼伦贝尔人，硕士研究生，2019年于南华大学获得学士学位，主要从事深度学习、图像分类等方面的研究。E-mail：631540532@qq.com

宁春玉（1976—），女，吉林农安人，硕士，副教授，硕士生导师，2005年于吉林大学获得计算机应用技术硕士学位，主要从事图像处理、模式识别的研究。E-mail：yeningcy@163.com

中图分类号: TP391.4
计量
- 文章访问数: 596
- HTML全文浏览量: 397
- PDF下载量: 234
- 被引次数: 0
出版历程
- 收稿日期: 2022-04-12
- 修回日期: 2022-05-03
- 录用日期: 2022-08-24
- 网络出版日期: 2022-08-24

Classification model based on fusion of multi-scale feature and channel feature for benign and malignant brain tumors

School of Life Science and Technology, Changchun University of Science and Technology, Changchun 130022, China

Funds: Supported by the Science and Technology Development Project of Jilin Province (No. 20200404219YY)

More Information

Corresponding author: yeningcy@163.com

摘要

摘要:
针对脑肿瘤良恶性分类过程复杂、分类准确率不高等问题，提出了一种基于多尺度特征与通道特征融合的分类模型。该模型以ResNeXt网络为主干网络，首先，将基于空洞卷积的多尺度特征提取模块代替第一层卷积层，利用膨胀率获取不同感受野的图像信息，将全局特征与局部显著特征相结合；其次，添加通道注意力机制模块，融合特征通道信息，提高对肿瘤区域的关注度，降低对冗余信息的关注度；最后，采用学习率的线性衰减策略、图像的标签平滑策略以及基于医学图像的迁移学习策略的组合优化提高模型的学习能力和泛化能力。在BraTS2017和BraTS2019数据集中进行实验，准确率分别达到98.11%和98.72%。与经典模型和其他先进方法相比，该分类模型能够有效地减少分类过程的复杂度，提高脑肿瘤良恶性分类的准确率。
- 脑肿瘤 /
- 多尺度特征 /
- 通道注意力机制 /
- 深度学习
Abstract:
Aiming at the problems of complex and inaccurate classification of benign and malignant brain tumors, a classification model was proposed based on the fusion of multi-scale and channel features. The model used ResNeXt as the backbone network. First, the multi-scale feature extraction module based on dilated convolution was used to replace the first convolution layer, which can make full use of dilation rates to obtain the image information from different receptive fields, and combine the global features with significant subtle ones. Second, the channel attention mechanism module was added in the network to fuse the feature channel information in order to increase the attention to the tumor, and reduce the attention to redundant information. Finally, the combination optimization strategy, the MultiStepLR strategy of the learning rate, the label smoothing strategy and the transfer learning strategy on medical images were adopted to improve the learning and generalization abilities of the model. The experiments were carried out on BraTS2017 Dataset and BraTS2019 Dataset, and the classification accuracy were 98.11% and 98.72%, respectively. Compared with other advanced methods and classical models, the proposed classification model can effectively reduce the complexity of the classification process and improve the detection accuracy of benign and malignant brain tumors.
- brain tumor /
- multi-scale feature /
- channel attention mechanism /
- deep learning

HTML全文

图 1 MDCA-ResNeXt网络结构

Figure 1. MDCA-ResNeXt network structure

下载: 全尺寸图片幻灯片

图 2 $ C = 32 $的ResNeXt结构^[20]

Figure 2. ResNeXt structure with $ C = 32 $^[20]

下载: 全尺寸图片幻灯片

图 3 不同膨胀率的空洞卷积

Figure 3. Dilated convolution results with different dilation rates

下载: 全尺寸图片幻灯片

图 4 MD模块

Figure 4. MD module

下载: 全尺寸图片幻灯片

图 5 CA模块

Figure 5. CA module

下载: 全尺寸图片幻灯片

图 6 4种模态下的HGG图像

Figure 6. HGG images in four modalities

下载: 全尺寸图片幻灯片

图 7 4种模态下的LGG图像

Figure 7. LGG images in four modalities

下载: 全尺寸图片幻灯片

图 8 去除噪声前后对比图

Figure 8. Comparison before and after preprocessing

下载: 全尺寸图片幻灯片

图 9 3种网络的分类结果评价图

Figure 9. Evaluation diagram of classification results for three kinds of Nets

下载: 全尺寸图片幻灯片

图 10 HGG的原始图像和特征可视化图

Figure 10. Original image and feature visualizations of HGG

下载: 全尺寸图片幻灯片

图 11 LGG的原始图像和特征可视化图

Figure 11. Original image and feature visualization of LGG

下载: 全尺寸图片幻灯片

表 1 实验数据集分布

Table 1. Distribution of experimental datasets

数据集	肿瘤类别	数据分布		总数
数据集	肿瘤类别	训练集	测试集	总数
BraTS2017 数据集	HGG	840	210	1050
BraTS2017 数据集	LGG	900	225	1125
BraTS2019 数据集	HGG	1035	260	1295
BraTS2019 数据集	LGG	915	225	1140

下载: 导出CSV

表 2 优化前BraTS2017数据集的分类结果评价表

Table 2. Evaluation of classification results on BraTS2017 before optimization

网络	ACC（%）	SEN（%）	SPE（%）	PPV（%）	NPV（%）
ResNet	89.15±1.83	88.76±3.38	89.51±3.38	88.92±3.67	89.60±2.58
SENet	90.44±3.25	93.05±2.09	89.25±7.41	88.21±5.69	93.19±1.77
ResNeXt	90.34±1.14	89.23±3.36	92.53±2.18	91.85±1.96	90.22±2.57
MDCA-ResNeXt	93.19±0.35	93.05±1.67	93.33±1.69	92.91±1.54	93.54±1.36

下载: 导出CSV

表 3 优化前BraTS2019数据集的分类结果评价表

Table 3. Evaluation of classification results on BraTS2019 before optimization

网络	ACC（%）	SEN（%）	SPE（%）	PPV（%）	NPV（%）
ResNet	91.83±2.73	93.31±2.12	90.13±4.59	91.71±3.65	92.10±2.51
SENet	91.91±2.42	90.54±5.63	91.74±4.83	93.00±3.76	93.25±2.54
ResNeXt	93.57±1.50	94.23±2.02	92.80±2.99	93.85±2.31	93.35±2.15
MDCA-ResNeXt	94.10±1.40	94.38±1.67	93.78±2.33	94.76±2.12	93.60±1.59

下载: 导出CSV

表 4 优化后BraTS2017数据集的分类结果评价表

Table 4. Evaluation of classification results on BraTS2017 after optimization

网络	ACC（%）	SEN（%）	SPE（%）	PPV（%）	NPV（%）
Improved ResNet	96.87±1.49	96.76±0.71	96.98±2.90	96.84±2.97	96.98±0.64
Improved SENet	97.56±1.04	96.67±0.89	98.40±1.43	98.27±1.52	96.94±0.83
Improved ResNeXt	97.98±1.33	97.43±2.06	98.49±1.28	98.38±1.39	97.63±1.88
Improved MDCA-ResNeXt	98.11±0.41	97.43±0.26	98.76±0.91	98.66±0.97	97.63±0.22

下载: 导出CSV

表 5 优化后BraTS2019数据集的分类结果评价表

Table 5. Evaluation of classification results on BraTS2019 after optimization

网络	ACC（%）	SEN（%）	SPE（%）	PPV（%）	NPV（%）
Improved ResNet	97.03±1.95	97.31±2.02	96.62±3.04	97.20±2.55	96.91±2.23
Improved SENet	96.69±0.88	94.38±1.89	98.74±0.93	98.62±1.01	94.99±1.57
Improved ResNeXt	97.98±0.57	97.69±0.47	98.31±0.73	98.53±0.64	97.36±0.54
Improved MDCA-ResNeXt	98.72±0.31	98.62±0.64	98.85±0.51	99.00±0.44	98.41±0.73

下载: 导出CSV

表 6 先进方法分类结果对比表

Table 6. Comparison of classification results of advanced methods

文献	方法	肿瘤分割	数据集	准确率(%)
文献[7]	HCS+ Multi-SVNN	是	BraTs2014	93.00
文献[15]	Inception V3+POS	是	BraTs2017	96.90
文献[16]	VGG16+ELM	否	BraTs2017	96.90
文献[17]	3D CNN+VGG19+FNN	是	BraTs2017	96.97
文献[8]	FBSO	是	BraTs2018	93.85
文献[19]	3D U-Net	否	BraTs2018	91.67
本文方法	Improved MDCA-ResNeXt	否	BraTs2017	98.11
本文方法	Improved MDCA-ResNeXt	否	BraTs2019	98.72

下载: 导出CSV

参考文献(30)

[1]	BRAY F, FERLAY J, SOERJOMATARAM I, et al. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries[J]. CA:A Cancer Journal for Clinicians, 2018, 68(6): 394-424. doi: 10.3322/caac.21492
[2]	BRAY F, FERLAY J, SOERJOMATARAM I, et al. Erratum: global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries[J]. CA:A Cancer Journal for Clinicians, 2020, 70(4): 313.
[3]	BAKAS S, AKBARI H, SOTIRAS A, et al. Advancing the cancer genome atlas glioma MRI collections with expert segmentation labels and radiomic features[J]. Scientific Data, 2017, 4: 170117. doi: 10.1038/sdata.2017.117
[4]	MOHAN G, SUBASHINI M M. MRI based medical image analysis: survey on brain tumor grade classification[J]. Biomedical Signal Processing and Control, 2018, 39: 139-161. doi: 10.1016/j.bspc.2017.07.007
[5]	KHARRAT A, HALIMA M B, AYED M B. MRI brain tumor classification using Support Vector Machines and meta-heuristic method[C]. IEEE 2015 15th International Conference on Intelligent Systems Design and Applications (ISDA), IEEE, 2015: 446-451.
[6]	徐立, 白金牛, 李磊民. 基于脑部MR图像GMM特征决策分类的肿瘤诊断[J]. 控制工程,2017,24(8):1718-1722. doi: 10.14107/j.cnki.kzgc.150604 XU L, BAI J N, LI L M. Diagnosis of tumor in brain MR images based on GMM features and decision tree classifier[J]. Control Engineering of China, 2017, 24(8): 1718-1722. (in Chinese) doi: 10.14107/j.cnki.kzgc.150604
[7]	RAJU A R, SURESH P, RAO R R. Bayesian HCS-based multi-SVNN: a classification approach for brain tumor segmentation and classification using Bayesian fuzzy clustering[J]. Biocybernetics and Biomedical Engineering, 2018, 38(3): 646-660. doi: 10.1016/j.bbe.2018.05.001
[8]	NARMATHA C, ELJACK S M, TUKA A A R M, et al. . A hybrid fuzzy brain-storm optimization algorithm for the classification of brain tumor MRI images[J]. Journal of Ambient Intelligence and Humanized Computing,doi: 10.1007/s12652-020-02470-5.
[9]	黄乐弘, 曹立华, 李宁, 等. 深度学习的空间红外弱小目标状态感知方法[J]. 中国光学,2020,13(3):527-536. HUANG L H, CAO L H, LI N, et al. A state perception method for infrared dim and small targets with deep learning[J]. Chinese Optics, 2020, 13(3): 527-536. (in Chinese)
[10]	郑江鹏, 余平, 赵萌, 等. 利用低信噪比小样本太赫兹光谱实现心肌淀粉样变检测[J]. 中国光学,2022,15(3):443-453. doi: 10.37188/CO.2021-0223 ZHENG J P, YU P, ZHAO M, et al. Detection of myocardial amyloidosis by a small number of terahertz spectra with low signal-to-noise ratio[J]. Chinese Optics, 2022, 15(3): 443-453. (in Chinese) doi: 10.37188/CO.2021-0223
[11]	吴海滨, 魏喜盈, 王爱丽, 等. 八度卷积和双向门控循环单元结合的X光安检图像分类[J]. 中国光学,2020,13(5):1138-1146. doi: 10.37188/CO.2020-0073 WU H B, WEI X Y, WANG A L, et al. X-ray security inspection images classification combined octave convolution and bidirectional GRU[J]. Chinese Optics, 2020, 13(5): 1138-1146. (in Chinese) doi: 10.37188/CO.2020-0073
[12]	李宇, 刘雪莹, 张洪群, 等. 基于卷积神经网络的光学遥感图像检[J]. 光学精密工程,2018,26(1):200-207. doi: 10.3788/OPE.20182601.0200 LI Y, LIU X Y, ZHANG H Q, et al. Optical remote sensing image retrieval based on convolutional neural networks[J]. Optics and Precision Engineering, 2018, 26(1): 200-207. (in Chinese) doi: 10.3788/OPE.20182601.0200
[13]	陈筱, 朱向冰, 吴昌凡, 等. 基于迁移学习与特征融合的眼底图像分类[J]. 光学精密工程,2021,29(2):388-399. doi: 10.37188/OPE.20212902.0388 CHEN X, ZHU X B, WU CH F, et al. Research on fundus image classification based on transfer learning and feature fusion[J]. Optics and Precision Engineering, 2021, 29(2): 388-399. (in Chinese) doi: 10.37188/OPE.20212902.0388
[14]	GNANASEKARAN V S, JOYPAUL S, SUNDARAM P M, et al. Deep learning algorithm for breast masses classification in mammograms[J]. IET Image Processing, 2020, 14(12): 2860-2868. doi: 10.1049/iet-ipr.2020.0070
[15]	SHARIF M I, LI J P, KHAN M A, et al. Active deep neural network features selection for segmentation and recognition of brain tumors using MRI images[J]. Pattern Recognition Letters, 2020, 129: 181-189. doi: 10.1016/j.patrec.2019.11.019
[16]	KHAN M A, ASHRAF I, ALHAISONI M, et al. Multimodal brain tumor classification using deep learning and robust feature selection: a machine learning application for radiologists[J]. Diagnostics, 2020, 10(8): 565. doi: 10.3390/diagnostics10080565
[17]	REHMAN A, KHAN M A, SABA T, et al. Microscopic brain tumor detection and classification using 3D CNN and feature selection architecture[J]. Microscopy Research and Technique, 2021, 84(1): 133-149. doi: 10.1002/jemt.23597
[18]	SEETHA J, RAJA S S. Brain tumor classification using Convolutional Neural Networks[J]. Biomedical and Pharmacology Journal, 2018, 11(3): 1457-1461. doi: 10.13005/bpj/1511
[19]	赵尚义, 王远军. 基于3D CNN的脑胶质瘤分类算法[J]. 光学技术,2019,45(6):749-755. ZHAO SH Y, WANG Y J. Brain glioma classification algorithm based on 3D CNN[J]. Optical Technique, 2019, 45(6): 749-755. (in Chinese)
[20]	XIE S N, GIRSHICK R, DOLLÁR P, et al. . Aggregated residual transformations for deep neural networks[C]. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, 2017: 5987-5995.
[21]	SZEGEDY C, VANHOUCKE V, IOFFE S, et al. . Rethinking the inception architecture for computer vision[C]. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, 2016: 2818-2826.
[22]	HE K M, ZHANG X Y, REN SH Q, et al. . Deep residual learning for image recognition[C]. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, 2016: 770-778.
[23]	YU F, KOLTUN V. Multi-scale context aggregation by dilated convolutions[C]. 4th International Conference on Learning Representations, 2016: https://doi.org/10.48550/arXiv.1511.07122.
[24]	WOO S, PARK J, LEE J Y, et al. . CBAM: convolutional block attention module[C]. Proceedings of the 15th European Conference on Computer Vision, Springer, 2018: 3-19.
[25]	JIANG L Q, NING CH Y, LI J Y. Glioma classification framework based on SE-ResNeXt network and its optimization[J]. IET Image Processing, 2022, 16(2): 596-605. doi: 10.1049/ipr2.12374
[26]	CHENG J, HUANG W, CAO SH L, et al. Enhanced performance of brain tumor classification via tumor region augmentation and partition[J]. PLoS One, 2015, 10(12): e0140381.
[27]	VOVK U, PERNUS F, LIKAR B. A review of methods for correction of intensity inhomogeneity in MRI[J]. IEEE Transactions on Medical Imaging, 2007, 26(3): 405-421. doi: 10.1109/TMI.2006.891486
[28]	TUSTISON N J, AVANTS B B, COOK P A, et al. N4ITK: improved N3 bias correction[J]. IEEE Transactions on Medical Imaging, 2010, 29(6): 1310-1320. doi: 10.1109/TMI.2010.2046908
[29]	ZHONG ZH, ZHENG L, KANG G L, et al. . Random erasing data augmentation[C]. Proceedings of the 34th AAAI Conference on Artificial Intelligence, AAAI, 2020: 13001-13008.
[30]	HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, 2018: 7132-7141.