用于ROP计算机辅助筛查的广角视网膜图像DNN分类器

数据集获取（Figshare）

• 源站标题：DNN Classifier of Wide-angle Retinal Images in Computer-Aided Screening for ROP
• DOI：10.6084/m9.figshare.6367685.v1
• 发布日期：2018-06-05
• 文件数：6 个
• 文件总体积：1.56 GB
• 源站页面：https://figshare.com/articles/dataset/DNN_Classifier_of_Wide-angle_Retinal_Images_in_Computer-Aided_Screening_for_ROP/6367685

命令行下载

curl -L -o "data set preview.png" "https://ndownloader.figshare.com/files/11854148"
curl -L -o "Training Set Thumbnails.zip" "https://ndownloader.figshare.com/files/11854223"
curl -L -o "Test Set Thumbnails.zip" "https://ndownloader.figshare.com/files/11854229"
curl -L -o "DNNs.zip" "https://ndownloader.figshare.com/files/11694419"
curl -L -o "HumanExpertsPerformance.zip" "https://ndownloader.figshare.com/files/11854154"
curl -L -o "scripts.zip" "https://ndownloader.figshare.com/files/11854157"

文件清单

• data set preview.png（835.06 KB）
• Training Set Thumbnails.zip（35.05 MB）
• Test Set Thumbnails.zip（11.50 MB）
• DNNs.zip（1.51 GB）
• HumanExpertsPerformance.zip（254.70 KB）
• scripts.zip（21.16 KB）

文件总体积 1.56 GB（源站 API 实测）。

数据加载示例（图像类）

from PIL import Image
import glob, os

files = (glob.glob(os.path.join(path, "**", "*.png"), recursive=True)
       + glob.glob(os.path.join(path, "**", "*.jpg"), recursive=True)
       + glob.glob(os.path.join(path, "**", "*.tif"), recursive=True))
print("图像文件数:", len(files))
img = Image.open(files[0]); print("尺寸/模式:", img.size, img.mode)

# torchvision Dataset 方式:
# from torchvision import datasets
# ds = datasets.ImageFolder(path)  # 要求 子目录=类别

目录组织与标注格式以源站说明和下载后实际文件为准。

数据说明

This repository contains the following files:

1. data set preview.png

A preview of the data set. This is a snapshot of the resized (80x60) version of the training & test data set.
The original image data set are not provided, due to patient privacy protection and hospital IRB regulations.

2. Training Set Thumbnails.zip

The training dataset includes:
Class I (with disease): 8089
Class II (without disease): 9711

3. Test Set Thumbnails.zip

The test training dataset includes:
Class I (with disease): 155
Class II (without disease): 1587

4. DNNs.zip

This zip file contains 3 subfolders: AlexNet, GoogLeNet, VGG16.
Each subfolder further includes:
Model definition graph
Two trained models in the Caffe format. One is trained “from scratch” and the other “by transfer learning”.
P-R curve
ROC curve
Softmax output probabilities on the test data set

5. HumanExpertsPerformance.zip

The five pediatric ophthalmologists’’ diagnoses on the test data set
DNN performance measures
DNN vs Human P-R curve
DNN vs Human ROC curve

6. scripts.zip

Scripts used in this study for batch inference and data analysis.

许可

MIT

引用

学术使用请引用 DOI 10.6084/m9.figshare.6367685.v1（source: https://figshare.com/articles/dataset/DNN_Classifier_of_Wide-angle_Retinal_Images_in_Computer-Aided_Screening_for_ROP/6367685）。