增强Bagging集成学习及多目标检测算法

doi:10.13229/j.cnki.jdxbgxb20220503

摘要/Abstract

摘要：

针对现有目标检测算法在医疗影像疾病检测中存在定位精度和分类准确率较低的问题，提出了一种基于动态加权Bagging集成学习的多目标检测方法。以胸部影像疾病检测为例，引入联合注意力（CA）模块增强区域感受野，提升弱学习器对目标区域的定位能力；使用动态加权Bagging集成学习方法，根据置信度赋予弱学习器投票权重，降低了模型方差，改善了泛化误差，提升了分类准确率。实验结果表明，在胸部影像疾病检测任务中，本文算法的平均检测精度达到41.9%，相较于YOLOv5原型提高了2.5%，同时G-mean提升了1.3%；将模型加权集成后，平均检测准确率达到81.06%，相较于原模型提升了1.58%，具有较高的定位精度和分类准确率。因此，本文算法可以更好地完成胸部影像疾病检测任务。

关键词: 计算机应用, 胸部影像疾病检测, 动态加权Bagging集成学习, 联合注意力

Abstract:

Aiming at the problems of low localization accuracy and classification accuracy in medical imaging disease detection by existing target detection algorithms， a multi-target detection method based on dynamic weighted Bagging ensemble learning was proposed. Taking chest imaging disease detection as an example， the Coordinate Attention （CA） module was introduced to enhance the receptive field of the region and improve the weak learner's ability to locate the target region. The dynamic weighted Bagging ensemble learning method was used to give the weak learner voting weight according to the confidence. The variance of the model was reduced， the generalization error was improved， and the classification accuracy was improved. The experimental results show that in the chest imaging disease detection task， the average detection accuracy of the proposed algorithm reaches 41.9%， which is 2.5% higher than that of the YOLOv5 prototype， and G-mean is improved by 1.3%； after the model is weighted and integrated， the average detection accuracy rate reaches 81.06%， which is 1.58% higher than the original model， and has high positioning accuracy and classification accuracy. Therefore， the proposed algorithm can better complete the task of chest imaging disease detection.

Key words: computer application, chest imaging disease detection, dynamically weighted Bagging ensemble learning, coordinate attention

中图分类号:

TP391.4

车翔玖,于英杰,刘全乐. 增强Bagging集成学习及多目标检测算法[J]. 吉林大学学报(工学版), 2022, 52(12): 2916-2923.

Xiang-jiu CHE,Ying-jie YU,Quan-le LIU. Enhanced Bagging ensemble learning and multi⁃target detection algorithm[J]. Journal of Jilin University(Engineering and Technology Edition), 2022, 52(12): 2916-2923.

图/表 17

图1

图2

图3

表1

图4

图5

图6

图7

图8

表2

表3

表4

表5

表6

表7

图9

图10

参考文献 25

1	de Carvalho Filho A O, de Sampaio W B, Silva A C, et al. Automatic detection of solitary lung nodules using quality threshold clustering, genetic algorithm and diversity index[J]. Artificial Intelligence in Medicine, 2014, 60(3): 165-177.
2	Choi W J, Choi T S. Automated pulmonary nodule detection based on three-dimensional shape-based feature descriptor[J]. Computer Methods and Programs in Biomedicine, 2014, 113(1): 37-54.
3	Zhang Z, Zhang X, Lin X, et al. Ultrasonic diagnosis of breast nodules using modified faster R-CNN[J]. Ultrasonic Imaging, 2019, 41(6): 353-367.
4	Ramachandran S, George J, Skaria S, et al. Using YOLO based deep learning network for real time detection and localization of lung nodules from low dose CT scans[C]∥Medical Imaging 2018: Computer-Aided Diagnosis, Houston, Texas, USA, 2018, 10575: 347-355.
5	Johnson J M, Khoshgoftaar T M. Survey on deep learning with class imbalance[J]. Journal of Big Data, 2019, 6(1):1-54.
6	申彦, 朱玉全, 宋新平. 一种采用渐进学习模式的SBS-CLearning分类算法[J]. 江苏大学学报:自然科学版, 2018, 39(6): 696-703.
	Shen Yan, Zhu Yu-quan, Song Xin-ping. A classification learning algorithm of SBS-CLearning using step-by-step learning model[J]. Journal of Jiangsu University(Natural Science Edition), 2018, 39(6): 696-703.
7	Kazemi Y, Mirroshandel S A. A novel method for predicting kidney stone type using ensemble learning[J]. Artificial intelligence in Medicine, 2018, 84:117-126.
8	Farahani F V, Ahmadi A, Zarandi M H F. Lung nodule diagnosis from CT images based on ensemble learning[C]∥2015 IEEE Conference on Computational Intelligence in Bioinformatics and Computational Biology (CIBCB), Niagara Falls, ON, Canada, 2015:1-7.
9	Fitriyani N L, Syafrudin M, Alfian G, et al. Development of disease prediction model based on ensemble learning approach for diabetes and hypertension[J]. IEEE Access, 2019, 7:144777-144789.
10	Wang Yu-yan, Wang Du-juan, Geng Na, et al. Stacking-based ensemble learning of decision trees for interpretable prostate cancer detection[J]. Applied Soft Computing, 2019, 77: 188-204.
11	Zheng Hao, Zhang Yi-zhe, Yang Lin, et al. A new ensemble learning framework for 3D biomedical image segmentation[J]. AAAI Technical Track: Machine Learning, 2019, 33(1): 5909-5916.
12	Gao X Y, Amin A A, Shaban H H, et al. Improving the accuracy for analyzing heart diseases prediction based on the ensemble method[J]. Complexity, 2021, 2021:1-10.
13	Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Columbus, OH, USA, 2014:580-587.
14	Girshick R. Fast R-CNN[C]∥Proceedings of the IEEE International Conference on Computer Vision, Santiago, Chile, 2015:1440-1448.
15	Ren S Q, He K M, Girshick R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149.
16	牛杰, 钱堃, 卜雄洙. 基于YOLOv3算法和深度特征的地点识别方法[J]. 江苏大学学报:自然科学版, 2021, 42(6): 733-737.
	Niu Jie, Qian Kun, Bu Xiong-zhu. Place recognition method based on YOLOv3 and deep features[J]. Journal of Jiangsu University(Natural Science Edition), 2021, 42(6): 733-737.
17	Redmon J, Divvala S, Girshick R, et al. You only look once: Unified, real-time object detection[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, NV, USA, 2016:779-788.
18	Nguyen H Q, Lam K, Le L T, et al. VinDr-CXR: an open dataset of chest X-rays with radiologist's annotations[J/OL].[2022-04-15]. file:∥/C:/Users/ADMINI~1/AppData/Local/Temp/MicrosoftEdgeDownloads/4fb2e789-21c9-43c9-b06b-d06266b169fb/s41597-022-01498 -w.pdf
19	王进, 黄超, 王科, 等. 基于属性敏感卷积神经网络的行人属性分类[J]. 江苏大学学报:自然科学版, 2019, 40(4): 431-438.
	Wang Jin, Huang Chao, Wang Ke, et al. Pedestrian attribute classification based on label sensitive convolutional neural network[J]. Journal of Jiangsu University(Natural Science Edition), 2019, 40(4): 431-438.
20	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, UT, USA, 2018:7132-7141.
21	Woo S, Park J, Lee J Y, et al. CBAM: convolutional block attention module[J] Lecture Notes in Computer Science, 2018,11211: 3-19.
22	Hou Qin-bin, Zhou Da-quan, Feng Jia-shi. Coordinate attention for efficient mobile network design[C]∥2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Nashville, TN, USA, 2021:13713-13722.
23	袁梅, 全太锋, 黄俊, 等. 基于卷积神经网络的烟雾检测[J]. 重庆邮电大学学报: 自然科学版, 2020, 32(4): 620-629.
	Yuan Mei, Quan Tai-feng, Huang Jun, et al. Smoke detection based on convolutional neural network[J]. Journal of Chongqing University of Posts and Telecommunications (Natural Science Edition), 2020, 32(4): 620-629.
24	Kubat M, Matwin S. Addressing the curse of imbalanced training sets: one-sided selection[C]∥Proceedings of the Fourteenth International Conference on Machine Learning, Nashville, Tennessee, USA, 1997:179-186.
25	Breiman L. Bagging predictors[J]. Machine Learning, 1996, 24(2): 123-140.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

病理学标签	疾病描述	数量
主动脉扩张	发生在主动脉血管壁的异常隆起	3067
心脏扩大	成年病人的心胸比率大于0.5	2300
实质化	肺泡内充满液体、脓液、血液、细胞（包括肿瘤细胞）或其他物质，导致大叶性、弥漫性或多灶性不清混浊的病理过程。	386
胸膜增厚	涉及壁层或脏层胸膜的任何形式的增厚	1981
胸腔积液	胸腔内异常积液	1032
肺纤维化	肺里有多余的纤维组织	1617

模型	测试例1	测试例2	测试例3
1	√	ⅹ	ⅹ
2	ⅹ	√	ⅹ
3	ⅹ	ⅹ	√
集成	ⅹ	ⅹ	ⅹ

真实情况	预测结果
真实情况	正例	反例
正例	TP	FN
反例	FP	TN

方法	主动脉扩张	心脏扩大	实质化	胸膜增厚	胸腔积液	肺纤维化	均值
Faster-RCNN	0.521	0.830	0.537	0.434	0.443	0.354	0.520
YOLOv4	0.632	0.732	0.604	0.511	0.482	0.395	0.559
YOLOv5	0.651	0.800	0.635	0.532	0.500	0.452	0.595
CA-YOLOv5	0.662	0.828	0.641	0.538	0.524	0.464	0.610

方法	主动脉扩张	心脏扩大	实质化	胸膜增厚	胸腔积液	肺纤维化	均值
Faster-RCNN	0.500	0.545	0.202	0.185	0.283	0.160	0.313
YOLOv4	0.657	0.525	0.249	0.209	0.349	0.190	0.363
YOLOv5	0.672	0.539	0.263	0.222	0.402	0.265	0.394
CA-YOLOv5	0.699	0.550	0.323	0.235	0.428	0.278	0.419