SlideInstruction 和 SlideBench
包含最大WSI指令跟随数据集和病理图像多模态基准，用于训练和评估病理图像对话模型。
基本信息
资源简介
SlideInstruction是目前最大的全切片病理图像（WSI）指令跟随数据集，包含4,200张WSI标题和176,000个视觉问答对，覆盖多种病理类别。SlideBench是一个多模态基准，包括SlideBench-Caption、SlideBench-VQA (TCGA)和SlideBench-VQA (BCNB)，用于评估病理图像对话模型SlideChat在不同临床场景中的能力。数据模态为医学影像（WSI），任务涉及指令跟随和视觉问答，主要用于病理图像分析领域的模型训练与评估。
下载信息
注册下载
Tips: 该数据集需要在对应的数据源网站注册通过后，才能进行数据下载，注册有对应要求，或者需要收费。
暂未开放公开下载
Tips: 该数据集属于公开下载，应该可以免费公开下载。
免登录有偿下载
Tips: 该数据集 Qianfanghub 可以协助提供有偿下载服务，注意，服务不针对数据相关产权，只是技术服务费。
提供高速下载与技术交付服务（收技术服务费，非数据销售）
暂未开放千方医数集，医疗数据集部分，是为社区服务的公开医疗数据集搜索引擎，并不存储或者下载原始的任何数据。 如果您有其他医疗数据需求，可以和客服联系，或者下工单。我们有强大的三甲医疗机构帮助您提供个性化的医疗数据定制、采集、标注服务。
使用方式
数据集获取
git clone https://github.com/uni-medical/SlideChat.git
curl -L -o repo.zip https://github.com/uni-medical/SlideChat/archive/refs/heads/main.zip
unzip repo.zip
源站 README 摘录（使用方式）
SlideChat: A Multimodal Generative AI Assistant for Whole-Slide Computational Pathology
Note
This repository provides the official code for SlideChat. The expanded study, published in Nature Cancer, substantially extends our original CVPR 2025 work with a larger instruction dataset, broader expert-reviewed evaluation, and additional pathology tasks across cancer types.
Release
We release SlideChat, SlideInstruction, and SlideBench as open-source resources, hoping to facilitate research and development in computational pathology.
- SlideChat: The first large vision-language assistant for whole-slide pathology image analysis, capable of generating comprehensive descriptions and contextually relevant responses.
- SlideInstruction: The largest comprehensive WSI instruction-following dataset, derived from pathology reports…
- SlideBench: A WSI multimodal benchmark including SlideBench-Caption/Report (TCGA, CPTAC, HISTAI) and SlideBench-Closed(VQA-TCGA, VQA-BCNB, VQA-CPTAC, VQA-HISTAI).
Before open‑sourcing, SlideBench‑VQA‑TCGA underwent a second round of expert review with pathologists, further enhancing data quality. The initial version covers 10 cancer types with 1,494 samples. We later expanded it to 31 additional cancer types, rigorously validated by experts, yielding 3,176 samples (SlideBench‑VQA‑TCGA.csv). SlideChat achieved an accuracy of 75.23% on the initial version and 74.12% on the expanded version.
The results below correspond to the expanded journal version of SlideChat.
Closed-Ended Question Answering
| Dataset | Task | SlideChat | GPT-4o | MedDr | Quilt-LLaVA | LLaVA-Med |
|---|---|---|---|---|---|---|
| TCGA | Histopathological Changes | 0.859 ± 0.019 | 0.481 ± 0.027 | 0.533 ± 0.027 | 0.311 ± 0.025 | 0.292 ± 0.025 |
| TCGA | Cytomorphological Characteristics | 0.843 ± 0.040 | 0.584 ± 0.052 | 0.690 ± 0.050 | 0.180 ± 0.042 | 0.216 ± 0.045 |
| TCGA | Tissue Architecture | 0.836 ± 0.019 | 0.614 ± 0.026 | 0.601 ± 0.025 | 0.385 ± 0.026 | 0.465 ± 0.026 |
| TCGA | Tumor Characteristics | 0.717 ± 0.035 | 0.607 ± 0.039 | 0.612 ± 0.039 | 0.259 ± 0.034 | 0.306 ± 0.035 |
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。
精度瓶颈？数据缺失？
当前公开数据无法满足您的算法精度？千方提供针对 癌症（总论） 的高质量、多模态真实临床数据定制解决方案。
