医学图像
文章平均质量分 70
fanqiliang630
这个作者很懒,什么都没留下…
展开
-
pytorch lightning使用(简要介绍)
0. 简介pytorch lightning通过提供LightningModule和LightningDataModule,使得在用pytorch编写网络模型时,加载数据、分割数据集、训练、验证、测试、计算指标的代码全部都能很好的组织起来,显得主程序调用时,代码简洁可读性大幅度提升。1. pytorch lightning的安装pip install pytorch-lightningconda install pytorch-lightning -c conda-forge2. 定义一原创 2021-02-21 18:13:02 · 6539 阅读 · 5 评论 -
Attention gated networks: Learning to leverage salient regions in medical images
参考论文:“Attention gated networks : Learning to leverage salient regions in medical images”关键字:Attention Gate(AG) – automatically learns to focus on target structures主动学习将注意力放在关注的物体上的网络结构1. IntroductionAGs automatically learn to focus on target structu.原创 2021-01-10 22:25:56 · 2962 阅读 · 0 评论 -
Seeded Region Growing
参考论文:《Seeded Region Growing》 Rolf Adams and Leanne Bischof1. 算法描述:使用一系列的点(即所说的seed, 种子)来对图片进行分割初始时,将一系列的种子分组为n个集合,A1,A2,...,AnA_1, A_2, ..., A_nA1,A2,...,An 有时单独个一个集合只会包含一个点,需要去判断这是一个有效的特征(feature of interest)还是一个噪点(noise)随后SRG算法,将image进行分割,.原创 2020-12-20 22:29:19 · 1124 阅读 · 0 评论 -
Semi-Supervised Multi-Task Learning for Lung Cancer Diagnosis
参考论文:《Semi-supervised multi-task learning for lung cancer diagnosis》关键词:多任务学习(Multi-task learning, MTL),假阳性去除(FP reduction)1. 贡献:这是第一篇提出通过3D CNN进行分割和假阳性去除交叉训练的论文提出了一种通过MTL方法改进CAD的可能性这篇工作会带来一个全新的视角:如果处理缺乏医学图像标注的问题2. 方法:提出的3D 多任务学习算法基于CNN,并且通过一些共.原创 2020-12-20 13:45:13 · 869 阅读 · 2 评论 -
Pulmonary nodule detection in CT scans with equivariant CNNs
Pulmonary nodule detection in CT scans with equivariant CNNs1. G-CNN2. G-CNN implementation:参考论文: Pulmonary nodule detection in CT scans with equivariant CNNshttps://doi.org/10.1016/j.media.2019.03.010G-CNN: 3D group convolution1. G-CNN可能一听到group原创 2020-12-07 13:39:34 · 815 阅读 · 0 评论 -
Unsupervised representation learning with deep convolutional generative adversarial networks.md
论文:《Unsupervised representation learning with deep convolutional generative adversarial networks》Contribution:We propose and evaluate a set of constraints on the architectural topology of Convolutional GANs that make them stable to train in most setting原创 2020-11-24 16:18:05 · 852 阅读 · 0 评论 -
Pyramid Convolution(PyConv)
参考论文:《Pyramidal Convolution: Rethinking Convolutional Neural Networks for Visual Recognition》论文链接: http://arxiv.org/abs/2006.115381. 金字塔卷积形式:核心是使用多个大小不同的卷积对input feature maps进行卷积操作标准的卷积操作中,每个卷积核的通道和input feature的通道一致。PyConv中,随着卷积核size增大,通道大小减少,此时需原创 2020-11-12 12:22:56 · 4340 阅读 · 7 评论 -
Convolution Arithmetic
参考论文:A guide to convolution arithmetic for deep learning论文地址:http://arxiv.org/abs/1603.07285等效卷积核大小计算:K=k+(k−1)∗(r−1) (k为原大小,K为等效大小, r 为 dilation) -> 无空洞时dilation为1注意: floor(x-1)//2 + 1 == x//2 <- 当且仅当 x为偶数! 这种padding出现在ResNet中的downsample中1. .原创 2020-11-12 12:21:35 · 789 阅读 · 0 评论 -
Integrating lung parenchyma segmentation and nodule detection with deep multi-task learning
标题: Integrating lung parenchyma segmentation and nodule detection with deep multi-task learning参考论文:同标题1. 3D-Unet2. PDCB(Pyramid dilated convolution block)选取4种dilation rates,每种dilation rates对应一系列的卷积每一条卷积操作中,进行类似pyramid convolution(详情见:医学图像处理/2020/1.原创 2020-11-12 12:20:51 · 933 阅读 · 0 评论 -
An attempt at beating the 3D U-Net
参考论文: 《An attempt at beating the 3D U-Net》1. 实现的网络3D plain U-Netresidual 3D U-Netpre-acitvation residual 3D U-Net Identity mappings in deep residual networks.md这里逆卷积没有标识size,根据公式可以推出:stride=2, padding=0, kernel=(2, 2, 2)...原创 2020-11-12 12:19:48 · 999 阅读 · 0 评论 -
Identity mappings in deep residual networks
Identity mappings in deep residual networksIdentity mappings in deep residual networkspre-activation ResNetthe optimization is further eased because f is an identity mappingusing BN as pre-activation improves regularization of the models原创 2020-11-12 12:19:02 · 705 阅读 · 0 评论 -
Deep residual learning for image recognition
Deep residual learning for image recognition参考论文:《Deep residual learning for image recognition》pytorch 源码:https://pytorch.org/docs/stable/_modules/torchvision/models/resnet.html细节:We adopt batch normalization(BN) right after each convolution and原创 2020-11-12 12:18:09 · 701 阅读 · 0 评论 -
U-net:Convolutional networks for biomedical image segmentation
U-net:Convolutional networks for biomedical image segmentation参考论文:《U-net: Convolutional networks for biomedical image segmentation》https://lmb.informatik.uni-freiburg.de/people/ronneber/u-net/| max pooling | stride=2 || others | stride=1 (unpaddie原创 2020-11-12 12:16:12 · 851 阅读 · 0 评论