Skin-Path: A Vision-Language Dataset for Skin Cancer Pathology

A high-quality vision-language dataset for advancing AI-driven skin pathology research.

Abstract

Existing pathology VL datasets (e.g., PathCap, OpenPath, QUILT) suffer from low image quality, poor text-image alignment, and limited scalability, especially in skin cancer pathology. To bridge this gap, we introduce Skin-Path, a high-quality VL dataset curated from 194 H&E-stained WSIs at 20× magnification, with 277,761 image patches (300×300 px) and expert-annotated captions. Covering 10 skin diseases (e.g., seborrhoeic keratosis, basal cell carcinoma, squamous cell carcinoma), Skin-Path enables VL model training, medical report generation, and disease classification.

Research Objectives

Figure 1(a): Examples of WSI with associated captions and patches extracted at ×20 magnification.

Figure 1(b): Example of basal cell carcinoma (BCC) slide with associated captions and patches.

The primary goal of the Skin-Path dataset is to provide a high-quality vision-language dataset for advancing AI-driven pathology research, particularly in skin cancer diagnosis and histopathological analysis. Specific objectives include:

Enhance VL model performance in histopathology.
Overcome text-image misalignment in existing pathology datasets.
Support automated medical report generation with expert-verified captions.
Facilitate skin disease classification across 10 distinct conditions.
Encourage multimodal learning research in medical AI applications.

Dataset Details

Images: 194 whole slide images (WSIs) at 20× magnification.
Patches: 277,761 image patches (300×300 px).
Annotations: Pathologist-verified captions.
Diseases Covered: 10 types, including basal cell carcinoma & squamous cell carcinoma.

Dataset Access

Download: The dataset demo can be downloaded at Google Drive Link . The full version will be available soon.

Extract into:


Skin-Path/
├── Images/     # Image patches
├── Captions/   # Corresponding diagnostic reports

Word Cloud Representation

Figure 2: Word cloud of the Skin-Path dataset.

Terms of Use

Research-only: Not for commercial use without permission.
Ethical compliance: Required for medical AI applications.
Proper attribution: Cite the dataset in any publications.
No redistribution: Refer others to the official page.