论文阅读笔记:DeepRadiologyNet: Radiologist Level Pathology Detection in CT Head Images
2017-12-13 23:57
603 查看
简述
DeepRadiologyNet: Radiologist Level Pathology Detection in CT Head Images,深度放射科医学网络:基于放射医生或研究者水平的CT脑部头像的病理检测。该篇论文实现对CT脑部图像进行检测,自动检测出疾病。
模型
数据集
训练集约350万CT脑部图像,定义了约30个可观测的较严重的疾病标签。测试集约480万CT脑部图像,且与训练集图像不相交。
网络架构
公式
输入
训练集图片数量为N,图片宽度m,高度n,通道为L(L张不同灰度值的图片)。X=N∗m∗n∗L
标签
标签共K个,分别为0,1,2,…,K-1。Y={0,1,2,...,K−1}
分类概率判别式
ϕω(x)[k]表示输入x在参数ω下输出y=k的概率,即该图片属于某一类的概率。ϕω(x)[k]=P(y=k|x)
分类函数
f(x)作用是找到K个ϕω(x)[k]中最大概率的值k,即该图片属于哪一类。f:X→Y;x→y
f(x)=argmaxkϕω(x)[k]
损失函数
ω̂ 为损失函数,累加计算l(xi,yi)得到。ϕω(xi)[yi]表示图片xi属于标签yi的概率。l(xi,yi)=−logϕω(xi)[yi]
ω̂ =argmin∑(xi,yi)∼P(x,y)lω(xi,yi)
技术
DeepRadiologyNet采用技术
卷积神经网络采用类似GoogleNet的神经网络,每次训练时参数随机初始化,且训练集顺序随机打乱。损失函数loss采用多样的如Hinge loss、Soft-max logistic loss和Hierarchical loss。梯度更新采用Adam optimization 和Stochastic gradient descent with momentum(随机动量梯度下降)。Data Annotation(数据标注)
训练集数据标签由放射医学专家标注,考虑到数据集的规模太大,采用了相关数据库技术。(详细未知)总结
这篇论文采用的是常规的经典的图像分类方法,但是数据集很大,从不同渠道获取,较麻烦,但是系统表现良好,网络架构也较容易理解。Hierarchical loss目前不太理解,等请教导师后再补充。相关文章推荐
- 【论文阅读笔记】CheXNet: Radiologist-Level Pneumonia Detection on Chest X-Rays with Deep Learning
- 论文阅读(Zhuoyao Zhong——【aixiv2016】DeepText A Unified Framework for Text Proposal Generation and Text Detection in Natural Images)
- 《Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification》阅读笔记与实现
- Feed Forward and Backward Run in Deep Convolution Neural Network 论文阅读笔记
- 图割论文阅读笔记:Interactive Graph Cuts for Optimal Boundary & Region Segmentation of Objects in N-D Images
- 【论文阅读笔记】Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition
- 【论文阅读笔记】Deep Learning in Medical Imaging: Overview and Future Promise of an Exciting New Technique
- 论文笔记:A Global Covariance Descriptor for Nuclear Atypia Scoring in Breast Histopathology Images
- 论文笔记:Structure Inference Net: Object Detection Using Scene-Level Context and Instance-Level Relationships
- [论文解读] Deep Sliding Shapes for Amodal 3D Object Detection in RGB-D Images
- Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition--SPP-net论文笔记
- 论文阅读笔记:Threat of Adversarial Attacks on Deep Learning in Computer Vision: A Survey
- 《DropoutNet: Addressing Cold Start in Recommender Systems》论文阅读笔记
- 【翻译+原创】DeepFace: Closing the Gap to Human-Level Performance in Face Verification 论文笔记
- 论文阅读:Segmentation of teeth in CT volumetric dataset by panoramic projection and variational level se
- 论文阅读:Automatic Detection and Classication of Teeth in CT Data
- FaceBook 论文:DeepFace: Closing the Gap to Human-Level Performance in Face Verification 笔记
- Synthetic Data for Text Localisation in Natural Images论文阅读
- 论文笔记——SQUEEZENET ALEXNET-LEVEL ACCURACY WITH 50X FEWER PARAMETERS AND <0.5MB MODEL SIZE
- 论文阅读(Xiang Bai——【CVPR2015】Symmetry-Based Text Line Detection in Natural Scenes)