## 3D Context Enhanced Region-based Convolutional Neural Network (3DCE)
Developed by Ke Yan ([email protected], [yanke23.com](http://yanke23.com)), Imaging Biomarkers and Computer-Aided Diagnosis Laboratory, National Institutes of Health Clinical Center
3DCE [1] is an object detection framework which makes use of the 3D context in volumetric image data (and maybe video data) efficiently.
It was primarily designed for lesion detection in 3D CT images. However, the project also contains 2D Faster RCNN and R-FCN, which can be used for other object detection tasks.
Adapted from the code in [https://github.com/sxjscience/mx-rcnn-1](https://github.com/sxjscience/mx-rcnn-1)
## Introduction
* Implemented frameworks: Faster RCNN, R-FCN, Improved R-FCN [1], 3DCE R-FCN (see rcnn/symbol/symbol_vgg.py and tools/train.py)
* For the **DeepLesion** dataset [2,3,4], we:
* Load data split and annotations from DL_info.csv (see dataset/DeepLesion.py)
* Load images from 16-bit png files (see fio/load_ct_img.py)
* To preprocess the CT images, we: (see fio/load_ct_img.py)
* Linearly interpolate intermediate slices according to the slice interval
* Do intensity windowing
* Normalize pixel spacing
* Clip the black borders
* Other useful features:
* We evaluate on the validation set after each epoch. After several epochs, we evaluate the test set using the best model (see tools/train.py, validate.py, test.py, and core/tester.py)
* Adjustable batch size (num of images per batch) and iter_size (accumulate gradients in multiple iterations)
* Previous snapshots can be resumed by simply setting "exp_name" and "begin_epoch" in default.yml
* When running train.sh, it will generate log files named with "exp_name"
* Images can be prefetched from hard disk to speed up
##
#### Requirements
* MXNet 1.0.0
* Python 2.7
* Before running, run "make" to compile binary files
* To train the universal lesion detector, download the DeepLesion dataset [2]
#### File structure
* experiment_logs: log files for the results in our paper [1].
* images: images used in this readme.
* rcnn: the core codes. The main function is in core/tools/train.py.
* config.yml and default.yml: configuration files to run the code.
* train.sh and test.sh: run these files to train or test.
#### Notes
* To change dataset, implement your own data code according to DeepLesion.py and pascal_voc.py, and maybe change the data layer in core/loader.py.
* Only end-to-end training is considered in this project.
#### References
1. K. Yan, M. Bagheri, and R. M. Summers, “3D Context Enhanced Region-based Convolutional Neural Network for End-to-End Lesion Detection,” in MICCAI, 2018 ([arXiv](https://arxiv.org/abs/1806.09648))
1. The DeepLesion dataset. ([download](https://nihcc.box.com/v/DeepLesion))
1. K. Yan, X. Wang, L. Lu, and R. M. Summers, “DeepLesion: Automated Mining of Large-Scale Lesion Annotations and Universal Lesion Detection with Deep Learning,” J. Med. Imaging, 2018. ([paper](http://yanke23.com/papers/18_JMI_DeepLesion.pdf))
1. K. Yan et al., “Deep Lesion Graphs in the Wild: Relationship Learning and Organization of Significant Radiology Image Findings in a Diverse Large-scale Lesion Database,” in CVPR, 2018. ([arXiv](https://arxiv.org/abs/1711.10535))
![3DCE framework](images/3dce_framework.png)
![lesion detection results](images/3DCE_lesion_detection_results.png)
没有合适的资源?快使用搜索试试~ 我知道了~
温馨提示
本项目是医学研究的demo,内含数据集和教程 基于区域的卷积神经网络(Region-based Convolutional Neural Networks, RCNN)是一种用于目标检测的深度学习架构。它通过将卷积神经网络(CNN)与区域提议方法相结合,来识别图像中的特定区域,并在这些区域上应用卷积网络进行特征提取和分类。3D上下文增强则是指在处理三维数据时,考虑到的物体在空间中的位置、相邻物体的关系以及场景的整体结构等信息。
资源推荐
资源详情
资源评论
收起资源包目录
基于区域的卷积神经网络3D上下文增强.zip (122个子文件)
log.traintest_07-20_10-25-56.3DCE 1 image 92KB
log.traintest_07-01_13-46-14.3DCE 1 image 39KB
log.traintest_07-24_09-30-39.3DCE 1 image 1 slice 93KB
log.traintest_07-24_16-51-57.3DCE 1 image 11 slice 92KB
log.traintest_07-20_15-39-06.3DCE 3 image rerun 91KB
log.traintest_07-20_15-40-28.3DCE 5 image 91KB
log.traintest_07-20_15-42-03.3DCE 7 image 170KB
log.traintest_07-20_15-43-24.3DCE 9 image 170KB
_mask.c 689KB
maskApi.c 8KB
nms_kernel.cu 5KB
log.traintest_07-20_10-22-50.Faster RCNN 92KB
.gitignore 15B
maskApi.h 2KB
gpu_nms.hpp 148B
LEGAL NOTICE 846B
Makefile 263B
readme.md 3KB
3DCE_lesion_detection_results.png 598KB
3dce_framework.png 519KB
cocoeval.py 24KB
coco.py 18KB
module.py 17KB
loader.py 17KB
symbol_vgg.py 13KB
imdb.py 13KB
tester.py 13KB
train.py 13KB
load_ct_img.py 12KB
pascal_voc.py 11KB
proposal.py 10KB
rpn.py 10KB
rcnn.py 7KB
DeepLesion.py 6KB
pascal_voc_eval.py 6KB
setup.py 6KB
image.py 5KB
bbox_transform.py 5KB
config.py 5KB
bbox_regression.py 5KB
proposal_target.py 5KB
test.py 5KB
mask.py 5KB
metric.py 5KB
callback.py 3KB
evaluation.py 3KB
load_model.py 2KB
generate_anchor.py 2KB
draw_loss.py 2KB
load_data.py 2KB
nms.py 2KB
validate.py 2KB
save_model.py 762B
combine_model.py 709B
timer.py 700B
setup.py 599B
ds_utils.py 457B
_init_paths.py 215B
logger.py 119B
__init__.py 117B
__init__.py 57B
__init__.py 22B
__init__.py 0B
__init__.py 0B
__init__.py 0B
__init__.py 0B
__init__.py 0B
__init__.py 0B
__init__.py 0B
cocoeval.pyc 18KB
module.pyc 16KB
loader.pyc 16KB
coco.pyc 15KB
tester.pyc 12KB
imdb.pyc 12KB
pascal_voc.pyc 11KB
symbol_vgg.pyc 11KB
load_ct_img.pyc 10KB
proposal.pyc 8KB
rpn.pyc 8KB
bbox_transform.pyc 6KB
metric.pyc 6KB
proposal_target.pyc 6KB
rcnn.pyc 6KB
image.pyc 6KB
pascal_voc_eval.pyc 5KB
DeepLesion.pyc 5KB
test.pyc 5KB
bbox_regression.pyc 5KB
config.pyc 4KB
evaluation.pyc 4KB
callback.pyc 4KB
nms.pyc 3KB
generate_anchor.pyc 3KB
load_model.pyc 3KB
load_data.pyc 2KB
validate.pyc 2KB
mask.pyc 1KB
timer.pyc 1KB
ds_utils.pyc 1KB
共 122 条
- 1
- 2
资源评论
小码蚁.
- 粉丝: 2519
- 资源: 4067
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
最新资源
- Unity 插件之移动端影子生成插件(Mobile Fast Shadow 1.0.6)
- A.E.003.102 平面下客按钮(横标)B11E02-3818520F A1.dwg
- DoTween-PhotoWall-0701.unitypackage unity动态照片墙
- MT4946-VB一款2个N-Channel沟道SOP8的MOSFET晶体管参数介绍与应用说明
- 使用Unity开发的跳一跳小游戏
- OpenCV for Unity 2.6.0.unitypackage
- Unity 资源插件之僵尸模型资源包
- 0515student的一对多 多对多关系项目包
- 基于Hadoop的分布式并行增量爬虫技术研究.docx
- 基于Hadoop的海量数据存储平台设计与开发.docx
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功