MICCAI2019医学图像处理论文集：第6部分亮点

需积分: 9 110 浏览量更新于2024-07-16 收藏 228.74MB PDF 举报

"MICCAI2019-会议论文集第6部分 Part-VI.pdf，包含94篇在2019年医学图像处理顶级会议MICCAI上发表的论文，重点关注医学图像分析、分割、检测及重建技术。" 本文集中涵盖了多个关键领域的研究，以下是一些核心知识点的详细说明： 1. **多尺度粗到细分割方法**（Multi-Scale Coarse-to-Fine Segmentation for Screening Pancreatic Ductal Adenocarcinoma）：这篇文章探讨了一种用于胰腺导管腺癌筛查的多尺度粗到细分割技术。这种方法通过逐步细化的策略，提高了对肿瘤病灶的定位精度，特别适用于胰腺癌这类在图像中可能微小且难以识别的病变。 2. **MVP-Net：多视图FPN与位置感知注意力的深度通用病灶检测**（MVP-Net: Multi-view FPN with Position-aware Attention for Deep Universal Lesion Detection）： MVP-Net是一种融合了多视图信息和位置感知注意力机制的深度网络架构，旨在增强病灶检测的性能。它利用FPN（Feature Pyramid Network）提取不同尺度特征，并通过位置感知注意力机制来关注关键区域，从而提升对各种病灶的检测效果。 3. **空间频率非局部卷积LSTM网络**（Spatial-Frequency Non-Local Convolutional LSTM Network for pRCC classification）：这篇文章提出了一个用于肾细胞癌（pRCC）分类的空间频率非局部卷积LSTM网络。该模型结合了空间和频率域的信息，通过非局部操作捕捉全局上下文，同时利用LSTM处理序列数据，以提高分类准确性。 4. **BCD-Net：低剂量CT重建的加速、收敛和泛化**（BCD-Net for Low-dose CT Reconstruction: Acceleration, Convergence, and Generalization）： BCD-Net是为了解决低剂量CT图像的重建问题，通过加速算法、优化收敛性和增强泛化能力，提高图像质量和诊断的可靠性。这种方法有望减少辐射剂量，同时保持图像质量，对患者健康和临床诊断具有重要意义。 5. **双损失函数协作学习的腹部脂肪组织分割**（Abdominal Adipose Tissue Segmentation in MRI with Double Loss Function Collaborative Learning）：这项研究提出了一种结合双损失函数的协作学习方法，应用于磁共振成像（MRI）中的腹部脂肪组织分割。这种方法可以提高分割的准确性和稳定性，有助于肥胖、代谢疾病等的研究和诊断。以上五篇论文代表了2019年MICCAI会议中的部分研究成果，体现了当前医学图像处理领域的热点和创新方向，包括深度学习、多尺度分析、非局部操作、注意力机制以及优化的重建算法等。这些技术的进步对于提高医疗诊断的准确性和效率，以及推动精准医疗的发展都起到了重要作用。

Multi-Scale Coarse-to-Fine Segmentation for Screening PDAC 9

3. Dou, Q., Chen, H., Yu, L., Qin, J., Heng, P.A.: Multilevel contextual 3-d cnns for

false positive reduction in pulmonary nodule detection. IEEE TBE (2017)

4. Hussein, S., Chuquicusma, M.M., Kandel, P., Bolan, C.W., Wallace, M.B., Bagci,

U.: Supervised and unsupervised tumor characterization in the deep learning era.

arXiv:1801.03230 (2018)

5. Krizhevsky, A., Sutskever, I., Hinton, G.E.: Imagenet classiﬁcation with deep

convolutional neural networks. In: NIPS (2012)

6. Milletari, F., Navab, N., Ahmadi, S.A.: V-net: Fully convolutional neural networks

for volumetric medical image segmentation. In: 3DV (2016)

7. PDQ Adult Treatment Editorial Board: Pancreatic cancer treatment (PDQ

)

8. Ronneberger, O., Fischer, P., Brox, T.: U-net: Convolutional networks for biomed-

ical image segmentation. In: MICCAI (2015)

9. Roth, H.R., Lu, L., Farag, A., Shin, H.C., Liu, J., Turkbey, E.B., Summers,

R.M.: Deeporgan: Multi-level deep convolutional networks for automated pancreas

segmentation. In: MICCAI (2015)

10. Roth, H.R., Lu, L., Farag, A., Sohn, A., Summers, R.M.: Spatial aggregation of

holistically-nested networks for automated pancreas segmentation. MICCAI (2016)

11. Stewart, B.W.K.P., Wild, C.P., et al.: World cancer report 2014. Health (2017)

12. Xia, Y., Xie, L., Liu, F., Zhu, Z., Fishman, E.K., Yuille, A.L.: Bridging the gap

between 2d and 3d organ segmentation. MICCAI (2018)

13. Zhang, L., Lu, L., Summers, R.M., Kebebew, E., Yao, J.: Personalized pancreatic

tumor growth prediction via group learning. In: MICCAI (2017)

14. Zhou, Y., Xie, L., Fishman, E.K., Yuille, A.L.: Deep supervision for pancreatic

cyst segmentation in abdominal ct scans. In: MICCAI (2017)

15. Zhu, Z., Xia, Y., Shen, W., Fishman, E.K., Yuille, A.L.: A 3d coarse-to-ﬁne

framework for volumetric medical image segmentation. 3DV (2018)

MVP-Net: Multi-view FPN with Position-aware

Attention for Deep Universal Lesion Detection

Zihao Li

1,2∗

Shu Zhang

3∗

Junge Zhang

Kaiqi Huang

Yizhou Wang

3,2,4

Yizhou Yu

Institute of Automation, Chinese Academy of Sciences

Deepwise AI Lab

Computer Science Dept., Peking University

Peng Cheng Laboratory

Abstract. Universal lesion detection (ULD) on computed tomography

(CT) images is an important but underdeveloped problem. Recently,

deep learning-based approaches have been proposed for ULD, aiming

to learn representative features from annotated CT data. However, the

hunger for data of deep learning models and the scarcity of medical an-

notation hinders these approaches to advance further. In this paper, we

propose to incorporate domain knowledge in clinical practice into the

model design of universal lesion detectors. Speciﬁcally, as radiologists

tend to inspect multiple windows for an accurate diagnosis, we explicitly

model this process and propose a multi-view feature pyramid network

(FPN), where multi-view features are extracted from images rendered

with varied window widths and window levels; to eﬀectively combine

this multi-view information, we further propose a position-aware atten-

tion module. With the proposed model design, the data-hunger problem

is relieved as the learning task is made easier with the correctly induced

clinical practice prior. We show promising results with the proposed mod-

el, achieving an absolute gain of 5.65% (in the sensitivity of FPs@4.0)

over the previous state-of-the-art on the NIH DeepLesion dataset.

Keywords: Universal lesion detection· Multi-view · Position-aware · At-

tention.

1 Introduction

Automated detection of lesions from computed tomography (CT) scans can sig-

niﬁcantly boost the accuracy and eﬃciency of clinical diagnosis and disease

screening. However, existing computer aided diagnosis (CAD) systems usual-

ly focus on certain types of lesions, e.g. lung nodules [1], focal liver lesions [2],

thus their clinical usage is limited. Therefore, a Universal Lesion Detector which

can identify and localize diﬀerent types of lesions across the whole body all at

once is in urgent need.

Previous methods for ULD are largely inspired by the successful deep models

in the ﬁeld of natural images. For instance, Tang et al. [5] adapted a Mask-

RCNN [3] based approach to exploit the auxiliary supervision from manually

∗

indicates equal contribution. This work is done when Zihao Li is an intern at Deepwise

AI Lab.

2 Z. Li et al.

(a) [1024, 4096] (b) [50, 449] (c) [-505, 1980] (d) [446, 1960]

Fig. 1. CT images under diﬀerent window level and window width. (a) is the image

used in 3DCE. (b),(c),(d) are the multi-view images used in our MVP-Net.

generated pseudo mask. On the other hand, Yan et al. proposed a 3D Context

Enhanced (3DCE) RCNN model [6] which harness ImageNet pre-trained models

for 3D context modeling. Due to a certain degree of resemblance between natural

images and CT images, these advanced deep architectures also demonstrated

impressive results on ULD.

Nonetheless, the intrinsic quality of medical images should not be overlooked.

Beyond that, the inspection of medical images also exhibits diﬀerent character-

istics compared with recognition and detection of natural images. Therefore, it

would be helpful if we can eﬃciently exploit proper domain knowledge to devel-

op deep learning based diagnosis systems. We will try to analyze two aspects of

such domain knowledge, and explore how to formulate these human expertise

into a uniﬁed deep learning framework.

To accommodate for network input, previous studies [5, 6] use a signiﬁcant-

ly wide window

to compress CT’s 12bit Hounsﬁeld Uint (HU). However, this

would severely deteriorate the visibility of lesions as a result of degenerated im-

age contrast, as shown in Fig.1(a). In the clinical practice, fusing information

from multiple windows are eﬀective in improving the accuracy of detecting sub-

tle lesions and reducing false positives (FPs). During visual inspection of the CT

images, radiologists would combine complex information of diﬀerent inner struc-

tures and tissues from multiple reconstructions under diﬀerent window widths

and window levels to locate possible lesions. To imitate this process, we propose

to extract prominent features from three frequently examined window width-

s and window levels and capture complementary information across diﬀerent

windows with an attention based feature aggregation module.

During the inspection of whole body CT, body position of a slice (i.e. the

z-axis position of a certain slice), is also a frequently consulted prior knowledge.

Experienced specialists often rely on the underlying correspondence between

body position and lesion types to conduct lesion diagnosis. Moreover, radiologists

would use the position cue as an indicator for choosing proper window width and

window level. For instance, radiologists will mainly refer to the lung, bone and

mediastinal window when inspecting a chest CT. Therefore, it would be very

Windowing, also known as gray-level mapping, is used to change the appearance of the

picture to highlight particular structures.

4 Z. Li et al.

2 Methodology

Fig.2 gives an overview of the MVP-Net. For simplicity, we illustrate the case

that take three consecutive CT slices as network input. It should be noted that

MVP-Net can be easily extended to alternatives that take multiple slices as input

to consider 3D context like 3DCE [6].

The proposed MVP-Net takes three views of the original CT scan as in-

put and employs a late fusion strategy to fuse multi-view features before region

proposal network (RPN). As shown in part A of Fig.2, multi-view features are ex-

tracted from the three pathway backbone with shared parameters. Then, in part

B, to exploit the position information, a position recognition network is attached

to the concatenated multi-view features before RPN. Finally, a position-aware at-

tention module is further introduced to aggregate the multi-view features, which

will be passed to the RPN and RCNN network for the ﬁnal lesion detection. We

will elaborate these building blocks in the following subsections.

2.1 Multi-view FPN

The multi-view input for the MVP-Net is composed of multiple reconstructions

under diﬀerent window widths and window levels. Speciﬁcally, we adopt k-means

algorithm to cluster the recommended windows (labeled by radiologists) in the

DeepLesion dataset and obtain three most frequently inspected windows, whose

window levels and window widths are [50, 449], [−505, 1980] and [446, 1960] re-

spectively. As shown in Fig.1, these clustered windows approximately correspond

to the soft-tissue window, lung window, and the union of bone, brain, and me-

diastinal windows respectively.

As shown in Fig.2, we adopt a three pathway architecture to extract the

most prominent features from each representative view. FPN [4] is used as the

backbone network of each pathway. It takes in three consecutive slices as input

to model 3D context.

2.2 Attention based Feature Aggregation

Features extracted from diﬀerent views (windows) need to be properly aggregat-

ed for accurate diagnosis. A naive implementation for feature aggregation could

be concatenating them along the channel dimension. However, such an imple-

mentation would have to rely on the following convolution layers for eﬀective

feature fusion.

In the proposed MVP-Net, we employ a channel-wise attention based fea-

ture aggregation mechanism to adaptively reweight the feature maps of diﬀerent

views, imitating the process that radiologists put diﬀerent weights on multiple

windows for lesion identiﬁcation. We adopt an implementation similar to the

Convolutional Block Attention Module (CBAM) [9] to realize the channel-wise

attention. Details for the attention module is shown in Fig.2. Denoting F as

input feature map, we ﬁrstly aggregate the features with average pooling P

av g

and max pooling P

max

separately to extract representative descriptions, then

剩余843页未读，继续阅读

wujiankepj

粉丝: 4
资源: 5

MICCAI2019医学图像处理论文集：第6部分亮点

MICCAI2019-Part3.pdf

MICCAI2019-part2.pdf

MICCAI2018接收论文列表

MICCAI2019-会议论文集第5部分 Part-V.pdf

MICCAI2019-会议论文集 Part-IV.pdf

MICCAI2019-Contents.目录

MICCAI2019论文合集.txt

leetcode分类-MICCAI-AUTHOR-STATS:在这个项目中，我对MICCAI2010-2018的所有论文进行了调查，以寻找学者

miccai17-rnn-theano

MICCAI-2019.zip

最新资源