Skin Disease Classification Dataset
Description
Dermatological image dataset contains 1000 clinical images across 10 categories for skin disease and cancer classification. It includes malignant skin cancers, pre-cancerous lesions, benign conditions, and healthy controls. The dataset covers two malignant cancers: Melanoma and Squamous Cell Carcinoma. Melanoma is the most aggressive form of skin cancer and requires early detection. Squamous Cell Carcinoma is another common skin cancer that can metastasize if untreated. Actinic Keratosis represents pre-cancerous lesions that may progress to cancer. Benign conditions include Benign Keratosis (seborrheic keratoses), Dermatofibroma (benign fibrous nodules), and Melanocytic Nevus (common moles). These are non-cancerous but important for differential diagnosis. The dataset also includes inflammatory and infectious conditions: Atopic Dermatitis (chronic inflammatory condition) and Tinea Ringworm Candidiasis (fungal infections). Vascular Lesion covers various blood vessel abnormalities. A Healthy class provides normal skin samples as negative controls, essential for distinguishing pathological conditions from normal skin. This collection supports research in dermatological image analysis, enabling the development of classification systems that can distinguish between malignant cancers, pre-cancerous conditions, benign growths, inflammatory diseases, and healthy skin. The inclusion of both pathological and normal samples makes it suitable for binary and multi-class classification tasks in dermatology.
Files
Steps to reproduce
The Split_smol directory contains a preprocessed dermatological image dataset organized for classification tasks. It includes 878 clinical images across 10 categories: malignant skin cancers (Melanoma, Squamous Cell Carcinoma), pre-cancerous lesions (Actinic Keratosis), benign conditions (Benign Keratosis, Dermatofibroma, Melanocytic Nevus), inflammatory diseases (Atopic Dermatitis), infections (Tinea Ringworm Candidiasis), vascular abnormalities, and healthy skin controls. The directory is structured with separate train and val folders, each containing class-specific subdirectories with images. Three CSV files provide structured metadata: train.csv (697 entries), val.csv (181 entries), and classes.csv mapping class names to numeric IDs. Each CSV entry includes image paths, labels, and class identifiers, enabling direct integration with classification pipelines. The dataset supports binary and multi-class classification tasks in dermatological image analysis.