1. A deep feature fusion methodology for breast cancer diagnosis demonstrated on three imaging modality datasets_2017
方法: We present a methodology that extracts and pools low- to mid-level features using a pretrained CNN and fuses them with handcrafted radiomic features computed using conventional CADx methods.
Features are extracted from the five max-pooling layers, averagepooled across the channel (third) dimension, and normalized with L2 norm.?
沒(méi)啥用,利用VGG19進(jìn)行特征提取,讓后和傳統(tǒng)特征融合將每個(gè)pooling層的特征輸出1*1*dim維度的特征,使用了L2規(guī)則
2. A Method of Ultrasonic Image Recognition for Thyroid Papillary Carcinoma Based on Deep Convolution Neural Network_14 March 2018
該團(tuán)隊(duì)還有一篇文章基本一樣,數(shù)據(jù):307個(gè)人
the Fast Region-based Convolutional Network method (FasterRCNN) network is `improved and normalized by connecting the fourth layer and the fifth layer` of the shared convolution layer in the Faster RCNN network. Then, a multi-scale ultrasound image is used at the time of input.
將VGG16的第四層和第五層連接,同時(shí)將多尺度的超聲圖像輸入.
The experimental results show that compared with the original Faster RCNN network, the proposed method has higher recognition accuracy, shorter training time and higher efficiency in ultrasonic image recognition of thyroid papillary carcinoma.
與傳統(tǒng)的FasterRCNN相比這個(gè)方法更好,時(shí)間短,更精確
使用原始FasterRCNN在本文map = 0.6,原因就是超聲圖像各種不如自然圖像.
將層間連接之前使用L2 和尺度變化,然后輸入多尺度的圖像
Each tensor is normalized using L2, and normalization is accomplished within each pixel of the set feature tensor. After normalization, scaling is applied separately on each tensor.
Our experiments have proven that the feature in different size range can be learned through multi-scale image input, which increases the robustness, reduces the influence of down sampling on the feature representation, improves the extraction efficiency of the original feature of the image, and raises the accuracy of cancer feature recognition.
3.?A Region Based Convolutional Network for Tumor Detection and Classification in Breast Mammography_MICCAI_2016
The third step divides the images based on a grid representation to multiple overlapping sub images (parts) which are then used to train and test a modified Faster-RCNN
為了解決小尺度腫瘤和對(duì)比度低問(wèn)題將層間連接
features from lower levels of the CNN need to be taken into the account when making the decision as they are the only ones looking on the considered region proposals in the high enough resolution
類似三通道的FasterRCNN
4.?Automated Mass Detection in Mammograms Using Cascaded Deep Learning and Random Forests_2015
The first stage classifier consists of a multi-scale deep belief network that selects suspicious regions to be further processed by a two-level cascade of deep convolutional neural networks. The regions that survive this deep learning analysis are then processed by a two-level cascade of random forest classifiers that use morphological and texture features extracted from regions selected along the cascade. Finally, regions that survive the cascade of random forest classifiers are combined using connected component analysis to produce state-of-the-art results.
5.?Automatic Colon Polyp Detection Using Region Based Deep CNN and Post Learning Approaches_2018
We use a deep-CNN model (Inception Resnet) as a transfer learning scheme in the detection system.
將FasterRCNN的特征提取用inception resnet實(shí)現(xiàn),同時(shí)增加了一種尺度的anchor和更改閾值
Two versions of Inception Resnet have been introduced in [35] and we use a deeper version called Inception Resnet-v2.
6. Context-aware pedestrian detection especially for small-sized instances with Deconvolution Integrated Faster RCNN (DIF R-CNN)
解決行人檢測(cè)中的小目標(biāo)問(wèn)題,利用解卷積引入新的環(huán)境特征,能夠檢測(cè)像素＜50的人
Furthermore, the state-of-the-art CNN-based model (Inception-ResNet) is exploited to provide a rich and discriminative hierarchy of feature representations
Additionally, atrous convolution is adopted to enlarge the receptive field of the synthetic feature map
atrous convolution :稀疏卷積有洞的卷積
7.?Deep Learning for Automatic Detection of Abnormal Findings in Breast Mammography_2017
結(jié)構(gòu)也是三通道的FasterRCNN
8.?Large scale deep learning for computer aided detection of mammographic lesions_2016
將手動(dòng)選擇的特征與cnn的特征結(jié)合,得到cnn學(xué)不到的特征
In this paper we provide a head-to-head comparison between a state-of-the art in mammography CAD system, relying on a manually designed feature set and a Convolutional Neural Network (CNN)
9.?Liver Fibrosis Classification Based on Transfer Learning and FCNet for Ultrasound Images_2017
In this paper, we propose a novel liver fibrosis classification method based on transfer learning (TL) using VGGNet and a deep classifier called fully connected network (FCNet).
10.?R-FCN: Object Detection via Region-based Fully Convolutional Networks_2016
與FasterRCNN那種基于區(qū)域的方法不同本文用卷積的方法計(jì)算位置
用于公開數(shù)據(jù)集的自然圖像
Code is made publicly available at: https://github.com/daijifeng001/r-fcn.
our region-based detector is fully convolutional with almost all computation shared on the entire image. To achieve this goal, we propose position-sensitive score maps to address a dilemma between translation-invariance in image classification and translation-variance in object detection.
11.?Stacked deep polynomial network based representation learning for tumor classification with small ultrasound image dataset_2015
利用多項(xiàng)式網(wǎng)絡(luò)解決數(shù)據(jù)集過(guò)小的問(wèn)題
In this work, a stacked DPN (S-DPN) algorithm is proposed to further improv the representation performance of the original DPN, and S-DPN is then applied to the task of textur feature learning for ultrasound based tumor classification with small dataset.
