数据集获取

# 克隆仓库（含数据与代码）
git clone https://github.com/marcos-jimenez-larroy/Breast-cancer-detection-CNN.git

# 或仅下载打包文件
curl -L -o repo.zip https://github.com/marcos-jimenez-larroy/Breast-cancer-detection-CNN/archive/refs/heads/main.zip
unzip repo.zip

源站 README 摘录（使用方式）

Breast-cancer-detection-CNN

This repository includes a small python code where a CNN was trained on a small dataset containing images classified in 8 different categories; 4 of them representing Benign tumors and the other 4 malignant.

Motivation

Breast cancer is one of the most common cancer types suffered by women nowadays. In 2022 there were more than 2.3 million women diagnosed with it; and around 650 thousand deaths globally. This implied that any kind of mechanism we can construct to identify it early enough to start treatment would be a great advance. The most used technique for this is the use of mammographies (X-Ray image of the breast); which give us an image of the interior of the breast to see if there exists any harmful tumor. Sometimes this presence detection is not so straight-forward; so using Neural Networks could be a good solution; at least as a first filter so that real doctors can confirm or analyze in depth the most important results.

Breast density

The images used basically show breast density. If the breast contains high levels of fat; its density will be lower and any type of tumor (higher density) would be identified easier. However, is the breast has a noticeable density; the identification of a tumor may get harder. Moreover, women with denser breasts have a higher chance to get breast cancer than those with fatty breats; so it is definitely an important factor in breast cancer detection/diagnosys.

The Dataset

The dataset used here was taken from another GitHub repository (Breast-Caner-Detection) and contains about 5000 mammographies to be used as training/validation data (labeled) and about 1800 test images (which are NOT labeled). All images are in a (224,224,3) format. The labels used run from Density1 to Density4; increasing the breasts density as the number increases. They are also classified as Benign or Malignant.
Here you have some examples of the 8 different categories:

Link (TAKEN FROM THE MENTIONED REPOSITORY): https://drive.google.com/file/d/12umDKmXJ8ZmuiTrchSQRCs8SmRl12h/view

Image preprocessing

Before training the NN, we had a look at our images to see if they were detailed enough or if we could do some changes to them. We tried

数据说明

该数据集包含约5000张用于训练和验证的标记乳房X光图像，以及约1800张未标记的测试图像。所有图像均为(224,224,3)格式，标签从Density1到Density4表示乳房密度增加，并分为良性或恶性。用于乳腺癌检测和乳房密度分类任务。

完整仓库：github.com/marcos-jimenez-larroy/Breast-cancer-detection-CNN