BUET多中心阴道镜数据集

数据集获取

git clone https://github.com/mHealthBuet/BUET-colposcopy.git

curl -L -o repo.zip https://github.com/mHealthBuet/BUET-colposcopy/archive/refs/heads/main.zip
unzip repo.zip

源站 README 摘录（使用方式）

BUET Multi-Center Colposcopy Dataset

Overview

This repository contains a novel, multi-center colposcopy dataset created to address the limited representation of Bangladeshi patients in publicly available datasets. Curated in collaboration with multiple clinical sites in Dhaka, Bangladesh. This dataset provides a diverse and robust collection of clinical images and corresponding medical scores to aid in cervical cancer screening research.

Ethics & Data Privacy

Data collection and usage strictly adhered to ethical guidelines:

• IRB Approval: Protocols were approved by BUET and the respective collaborating clinical hospitals.
• Patient Privacy: A strict anonymization procedure was implemented to ensure no personally identifiable information (PII) was retained.
• Informed Consent: Participants explicitly provided informed consent and maintained the right to withdraw throughout the process.

Dataset Structure

After filtering for cases with adequate documentation, the combined dataset comprises 768 patient cases, yielding 3,072 images in total.
For each patient, a four-image series was obtained, corresponding to the following applications:

• Normal Saline
• Acetic Acid
• Green Filter
• Lugol’s Iodine
  Clinical Metadata Included:
• Transformation zone type
• Swede Score of five individual components (Aceto uptake,Margin, Vessel, Ledion Size, Iodine uptake)
• Total Swede score (which can be mapped to a provisional diagnosis)
• CIN Grading

Swede Score Distribution

The clinical reports document five distinct Swede Score characteristics. The score distributions across the dataset are summarized in the table below:

Characteristic	Score 0	Score 1	Score 2
Aceto-White Uptake	293	409	66
Margin and Surface	441	312	15
Vessels	635	84	49
Lesion Size	336	357	75
Iodine Staining	207	439	122

Image Directory Structure

The images are organized into patient-specific folders. Each folder (e.g., 001) corresponds to a single patient case and contains a four-image series sequentially numbered based on the application type:

image_folder/
├── 001/                  # Patient/Case ID
│   ├── 001.jpg           # Normal Saline applied image
│   ├── 002.jpg           # Green Filtered image
│   ├── 003.jpg           # Acetic Acid (A


## 数据加载示例（图像类）

```python
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。

完整仓库：github.com/mHealthBuet/BUET-colposcopy