SkinLex
SkinLex整合4个皮肤病数据集，含48种临床形态属性、20,411条图像记录，用于可解释皮肤病AI研究。
基本信息
资源简介
SkinLex是一个统一的皮肤病学AI可解释性数据集，整合了SkinCon、DermaCon-IN、MM-Skin和PASSION四个公开数据集的48种临床形态属性，共20,411条记录，涵盖临床、皮肤镜、病理等多种图像模态。数据集由巴塞尔大学等机构创建，用于概念瓶颈模型、特征相关性分析及可解释皮肤病诊断研究。
下载信息
注册下载
需要注册 Kaggle 账号并登录后下载，适合需要跟踪下载记录和使用 API 的用户。
暂未开放公开下载
无需注册即可直接获取公开样本或文档，适合快速预览和评估数据集质量。
免登录当前数据集主要来源为 Kaggle 公开托管，完整影像包建议通过原始链接或 Kaggle API 获取。
使用方式
数据集获取
git clone https://github.com/Digital-Dermatology/SkinLex.git
curl -L -o repo.zip https://github.com/Digital-Dermatology/SkinLex/archive/refs/heads/main.zip
unzip repo.zip
源站 README 摘录（使用方式）
SkinLex: Multi-Dataset Diagnostic Utility of Clinical Visual Concepts in AI Systems for Dermatology
This repository contains the poster and upcoming code release for the paper “Multi-Dataset Diagnostic Utility of Clinical Visual Concepts in AI Systems for Dermatology” (Accepted as Spotlight to the 11th ISIC Workshop @ MICCAI 2026).
The full code base, dataset preprocessing scripts, and LLM prompt templates are currently being prepared and will be made available here shortly.
Abstract & Overview
The clinical integration of AI decision-support systems in digital dermatology relies heavily on human trust. Clinical visual concepts act as an intermediate representation layer that can enhance human trust, corroborate model reliability, and enable intuitive auditing by medical staff.
SkinLex harmonizes 48 clinical morphological attributes across four public datasets (SkinCon, DermaCon-IN, MM-Skin, and PASSION) to create a unified benchmark of 20,411 records. Using this dataset, we evaluate how machine learning models utilize structured clinical concepts to predict 9-partition dermatological disease categories.
Key Insights
- Concept Complementarity: Restricting inputs to specific concept groups (e.g., colors or shapes alone) causes accuracy drops, showing that coarse diagnosis requires a varied combination of visual features.
- Concept Redundancy: Bootstrapped backward elimination reveals feature redundancy for algorithmic classification, allowing the omission of 8 concepts (for AP) and up to 18 concepts (for ROC-AUC) without significant performance loss.
- Clinical Utility: Identifying and removing redundant concepts could guide future concept set selection, potentially reducing annotation effort, improving modeling efficiency, and enhancing clinical trust.
Repository Contents
poster.pdf: PDF version of the poster presentation.code/: (Coming soon) Data harmonization pipelines, baseline models (LightGBM, Logistic Regression), and evaluation scripts.prompts/: (Coming soon) Qwen3.6 prompt templates used for diagnostic label mapping and concept extraction from text captions.
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。
需要用于商业化训练或医疗器械注册（NMPA/FDA）？
公开数据往往缺乏完整的病理回证与专家标注。千方医数集依托陕川双基地与超十家三甲医院直连，提供完全合规、去标识化、支持注册级质控的专病数据集交易服务。
