北京生物醫(yī)學(xué)工程

基于三維卷積經(jīng)網(wǎng)絡(luò)降低肺結(jié)節(jié)檢測假陽性的方法

Detection of false positive reduction of pulmonary nodules based on three-dimensional convolutional neural network

作者：侯智超楊楊李曉琴王曉曦高斌

單位：北京工業(yè)大學(xué)環(huán)境與生命學(xué)部(北京100124)， <p>通信作者:李曉琴。E-mail:[email protected]</p>

關(guān)鍵詞：圖像處理;深度學(xué)習(xí);三維卷積神經(jīng)網(wǎng)絡(luò);計(jì)算機(jī)輔助檢測;肺結(jié)節(jié)

分類號：R31804:TP391

出版年·卷·期（頁碼）：2022·41·4（352-359）

摘要：

目的為降低在計(jì)算機(jī)斷層掃描圖像中篩查肺結(jié)節(jié)的假陽性率，提出了一種基干三維卷積
神經(jīng)網(wǎng)絡(luò)降低肺結(jié)節(jié)檢測假陽性的方法。方法選用美國2016年肺結(jié)節(jié)分析挑戰(zhàn)賽提供的兩個(gè)版本的開源數(shù)據(jù)集，分別用干模型的訓(xùn)練和測試。首先應(yīng)用終像增強(qiáng)技術(shù)解決數(shù)據(jù)集中正負(fù)樣本分布不均衡的問題,并基于多視角采樣技術(shù)擴(kuò)充正樣本;基于自動編碼器及 K-means 無監(jiān)督聚類方法將負(fù)樣本分為5類并分別與正樣本組合得到了5個(gè)訓(xùn)練集該方法既減少了每個(gè)數(shù)據(jù)集中負(fù)樣本的樣本量又保證了負(fù)樣本的多樣性。然后搭建三維卷積神經(jīng)網(wǎng)絡(luò)，并分別使用構(gòu)建的5個(gè)訓(xùn)練集訓(xùn)練網(wǎng)絡(luò)，在此過程中不斷調(diào)整和優(yōu)化網(wǎng)絡(luò)結(jié)構(gòu)和參數(shù)得到5組降低肺結(jié)節(jié)假陽性檢測模型接著利用簡易集成法對肺結(jié)節(jié)進(jìn)行綜合判決。結(jié)果經(jīng)測試模型的敏感性和特異性分別為0.966和0.996通過FROC曲線計(jì)算得出 CPM得分為0.886。結(jié)論本文提出的方法可以有效降低肺結(jié)節(jié)檢測假陽性，可以為肺癌篩查工作提供有效幫助。

Objective In order to reduce the false-positive rate of pulmonarynodules in the preliminary screening of CT imagesa method was proposed based on the 3D convolutional neural network(3D-CNN) Methods Two versions of open source datasets provided by the 2016 pulmonary nodule analysis challenge in the United States were selected for training and testing of the modelTo balance the distribution of the nositive anc negative samples,we augmented the positive samples by the multi-view sampling and classified the negative samples into five categories bv the auto-encoder and the K-means unsupervised clustering. Then we got five training sets through combining the negative samples with the positive samples. This method reduced the size of negative samples and ensured its diversity. Next, we constructed the 3D convolutional neural network and respectively trained it with the five training sets,and adjusted the structure and parameters until obtaining a better model.Finallythe pulmonary nodules were judged by synthesizing the results of the five models.Results The sensitivity and specificity of the model were 0.966 and 0.996 respectively,and the CPM score was 0.886 calculated by the FROC curve.Conclusions The methoc we proposed can effectively reduce the false-positive detection of pulmonary nodules and provide help for lung
cancer screening.

參考文獻(xiàn)：

[1] Rolke HB,Bakke PS,Gallefoss F.Delays in the diagnostic
pathways for primary pulmonary carcinoma in Southern Norway J1.Respiratory Medicine,2007101(6):1251-1257.
[2] National Lung Screening Trial Research Team; Aberle DR,
Adams AM,Berg CD, et al.Reduced lung-cancer mortality with low-dose computed tomographic screening[J].The New England Journal of Medicine2011365(5):395-409.
[3]趙愉,魯雯,王遠(yuǎn)軍.肺結(jié)節(jié)計(jì)算機(jī)輔助檢測技術(shù)研究概述
J.生物醫(yī)學(xué)工程學(xué)雜志,2014,31(5):1172-1177.
Zhao Y,Lu W,Wang YJ.A review on the research progress of the computer-aided detection of pulmonarynodule[J]. Journal of Biomedical Engineering,201431(5):1172-1177
[4] Anirudh R,Thiagarajan JJ,Bremer Tet al.Lung nodule detection
using 3D convolutional neural networks trained on weakly labeled data[C]//Conference on Medical Imaging 2016:Computer Aided Diagnosis. San Diego,CAUSA: SPIE,2016,9785:32-38.
[5] Cheng JZ,Ni D,Chou YHet al.Computer-aided diagnosis with
deep learning architecture: applications to breast lesions in US images and pulmonary nodules in CTscans[J].Scientific Reports,2016,6:24454.
[6] Setio AAA,Ciompi F,Litjens G，et al.Pulmonary nodule
detection in ct images:false positive reduction using multi-view convolutional networks[J. IEEE Transactions on Medica Imaging,2016,35(5):1160-1169
[7] Ding J,Li A,Hu Z,et al.Accurate pulmonary nodule detection ir
computed tomography images using deep convolutional neura networksM//Medical Image Computing and Computer Assisted Intervention-MICCAI 2017.Lecture Notes in Computer Science Berlin:Springer,Cham,201710435:559-567
[8] Kim BC,Yoon JS,Choi JSet al.Multi-scale gradual integratior
CNN for false positive reduction in pulmonarynodule detectior[J].Neural Networks,2019,115:1-10
[9] Setio AAA,Traverso A,de Bel Tet al.Validation comparison
and combination of algorithms for automatic detection of pulmonary nodules in computed tomography images:the LUNA16 challenge[J1.Medical Image Analysis,2017,42:1-13
[10]" Tang H,Kim DR,Xie X.Automated pulmonary nodule detection
using 3D deep convolutional neural networks[C]//2018IEEE15th International Symposium on Biomedical Imaging Washington,DCUSA:IEEE Press2018:523-526
[11]張福玲,張少敏.應(yīng)用于CT圖像肺結(jié)節(jié)檢測的深度學(xué)習(xí)方法
綜述[J].計(jì)算機(jī)工程與應(yīng)用，202056(13):20-32
Zhang FL,Zhang SM.Review of deep learning methods applied to lung nodule detection in CT images[J].Computer Engineering and Applications,2020,56(13):20-32
「12] 謝未央,陳彥博,王季勇,等.基于卷積神經(jīng)網(wǎng)絡(luò)的 CT 圖像肺結(jié)節(jié)檢測J1.計(jì)算機(jī)工程與設(shè)計(jì)201940(12):3575-3581
Xie WY,Chen YB,Wang JY,et al.Detection of pulmonary nodules in CT images based on convolutional neural networksJ]. Computer Engineering and Design,2019,40(12):3575-3581.
[13] Dou Q,Chen H,YuL,et al.Multi-level contextual 3D CNNs for
false positive reduction in pulmonarynodule detection[J] IEEE Transactions on Biomedical Engineering，2016，64(7):1558-1567.
[14] Xiao Z,Du N,Geng Let al.Multi-scale heterogeneous 3D CNN
for false-positive reduction in pulmonary nodule detection,based on chest CT images[J].Applied Sciences,20199(16):3261.[15] 高慧明,趙涓涓,劉繼華,等.多尺度卷積神經(jīng)網(wǎng)絡(luò)用于肺結(jié)節(jié)
假陽性降低[J].計(jì)算機(jī)工程與設(shè)計(jì),2019,40(9):2718-2724.
Gao HMZhao JILiu JHet al.Multi-scale convolutional neural
network for pulmonary nodule false positive reduction [J] Computer Engineering and Design,201940(9):2718-2724.[16] 秦紅,王皓,魏曉超,等.安全的常數(shù)輪多用戶k-均值聚類計(jì)
算協(xié)議[J].計(jì)算機(jī)研究與發(fā)展,202057(10):2188-2200 Qin H,Wang H,Wei XC,et al.Secure constant-round multi-userk-means clustering protocol[J]. Journal of Computer Research and Development,202057(10):2188-2200.
[17] Lecun Y,Bottou L,Bengio Y，et al.Gradient-based learning
applied to document recognition[J]. Proceedings of the IEEE,1998,86(11):2278-2324.
[18] Nair V,Hinton GE. Rectified linear units improve restricted
Boltzmann machinesC// the 27th International Conference on Machine Learning.HaifaIsraelICML-102010:807-814.
[19] Sun L,Li W,Ning X,et al.Gradient-enhanced softmax for face
recognition[J].IEICE Transactions on Information and Systems,2020E103.D(5):1185-1189
[20]任敬謀,李曉琴.基于非結(jié)節(jié)自動分類的二維卷積網(wǎng)絡(luò)在肺結(jié)
節(jié)檢測假陽性減少中的應(yīng)用[J].北京生物醫(yī)學(xué)工程，2020.39(4):389-397.
Ren JMLi XO.The two-dimensional convolution network based on nonnodule automatic classification for reduction offalse positivity in pulmonary nodule detection[J].Beijing Biomedical Engineering,2020,39(4):389-397.

服務(wù)與反饋：

【文章下載】【加入收藏】

提示：您還未登錄，請登錄！點(diǎn)此登錄

51黑料吃瓜在线观看,51黑料官网|51黑料捷克街头搭讪_51黑料入口最新视频