A new 2.5D representation for lymph node detection using random sets of deep convolutional neural network observations.

A new 2.5D representation for lymph node detection using random sets of deep convolutional neural network observations.
复制标题

DOI:
10.1007/978-3-319-10404-1_65
复制
发表时间:
2014
期刊:
LECTURE NOTES IN ARTIFICIAL INTELLIGENCE
影响因子:
--
通讯作者:
Summers, Ronald M.
Summers, Ronald M.
中科院分区:
其他
文献类型:
--
作者:
Roth, Holger R.;Lu, Le;Seff, Ari;Cherry, Kevin M.;Hoffman, Joanne;Wang, Shijun;Liu, Jiamin;Turkbey, Evrim;Summers, Ronald M.

文献摘要

参考文献

被引文献

相似文献

自动淋巴结(LN)检测是一项重要的临床诊断任务,但由于计算机断层扫描(CT)中周围结构的低对比度以及它们的不同尺寸、姿态、形状和稀疏分布的位置而非常具有挑战性。最先进的研究表明,在3.1个假阳性/体积(FP/vol)时,性能范围为52.9%灵敏度,或60.9%,在6.1 FP/vol。纵隔LN,通过3D HAAR特征的一次增强。在本文中,我们首先操作一个初步的候选生成阶段,以高FP水平(每个患者约40个)为代价,达到约100%的灵敏度,以收获感兴趣的体积(VOI)。因此,我们的2.5D方法分解任何3D VOI通过重新扫描2D重新格式化的正交视图N次,通过缩放,随机平移和相对于VOI质心坐标的旋转。然后,这些随机视图用于训练深度卷积神经网络(CNN)分类器。在测试中,CNN被用来为所有N个随机视图分配LN概率,这些概率可以被简单地平均(作为一个集合)以计算每个VOI的最终分类概率。我们在两个数据集上验证了该方法:90个CT体积,388个纵隔淋巴结和86个患者,595个腹部淋巴结。我们在纵膈和腹部分别实现了3 FP/vol时70%/83%和6 FP/vol时84%/90%的灵敏度,这比以前的最先进工作有了显著提高。
Automated Lymph Node (LN) detection is an important clinical diagnostic task but very challenging due to the low contrast of surrounding structures in Computed Tomography (CT) and to their varying sizes, poses, shapes and sparsely distributed locations. State-of-the-art studies show the performance range of 52.9% sensitivity at 3.1 false-positives per volume (FP/vol.), or 60.9% at 6.1 FP/vol. for mediastinal LN, by one-shot boosting on 3D HAAR features. In this paper, we first operate a preliminary candidate generation stage, towards ~100% sensitivity at the cost of high FP levels (~40 per patient), to harvest volumes of interest (VOI). Our 2.5D approach consequently decomposes any 3D VOI by resampling 2D reformatted orthogonal views N times, via scale, random translations, and rotations with respect to the VOI centroid coordinates. These random views are then used to train a deep Convolutional Neural Network (CNN) classifier. In testing, the CNN is employed to assign LN probabilities for all N random views that can be simply averaged (as a set) to compute the final classification probability per VOI. We validate the approach on two datasets: 90 CT volumes with 388 mediastinal LNs and 86 patients with 595 abdominal LNs. We achieve sensitivities of 70%/83% at 3 FP/vol. and 84%/90% at 6 FP/vol. in mediastinum and abdomen respectively, which drastically improves over the previous state-of-the-art work.
DOI: 10.1109/42.974920
发表时间: 2001-12-01
影响因子: 10.6
作者:
Göktürk, SB;Tomasi, C;Napel, S
通讯作者: Napel, S
DOI: 10.1016/j.media.2011.05.005
发表时间: 2012-01-01
影响因子: 10.9
作者:
Feuerstein, Marco;Glocker, Ben;Mori, Kensaku
通讯作者: Mori, Kensaku
DOI: 10.1093/jnci/92.3.205
发表时间: 2000-02-02
期刊: JOURNAL OF THE NATIONAL CANCER INSTITUTE
影响因子: --
作者:
Therasse, P;Arbuck, SG;Gwyther, SG
通讯作者: Gwyther, SG
DOI: 10.1109/tmi.2011.2168234
发表时间: 2012-02-01
影响因子: 10.6
作者:
Barbu, Adrian;Suehling, Michael;Comaniciu, Dorin
通讯作者: Comaniciu, Dorin
DOI: 10.1016/j.media.2012.11.001
发表时间: 2013-02-01
影响因子: 10.9
作者:
Feulner, Johannes;Zhou, S. Kevin;Comaniciu, Dorin
通讯作者: Comaniciu, Dorin