README.md

metadata

license: cc-by-nc-sa-4.0
task_categories:
  - image-classification
tags:
  - dermatology
  - skin-cancer
  - medical-imaging
  - HAM10000
pretty_name: DermaLens Skin Cancer Dataset
size_categories:
  - 10K<n<100K
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
      - split: test
        path: data/test-*
dataset_info:
  features:
    - name: image
      dtype: image
    - name: image_id
      dtype: string
    - name: lesion_id
      dtype: string
    - name: dx
      dtype: string
    - name: dx_type
      dtype: string
    - name: age
      dtype: float64
    - name: sex
      dtype: string
    - name: localization
      dtype: string
  splits:
    - name: train
      num_bytes: 2650122851
      num_examples: 9577
    - name: validation
      num_bytes: 689647139
      num_examples: 2492
    - name: test
      num_bytes: 354617559
      num_examples: 1285
  download_size: 3693404553
  dataset_size: 3694387549

DermaLens Skin Cancer Dataset

This dataset repo documents the data pipeline used to train the DermaLens V3 skin cancer classification model.

Source Dataset

HAM10000 (Human Against Machine with 10000 training images) — accessed via marmal88/skin_cancer on HuggingFace.

from datasets import load_dataset
ds = load_dataset("marmal88/skin_cancer")

Dataset Statistics

Split	Images	Malignant	Benign	Positive Rate
Train	10,683	~2,093	~8,590	19.6%
Validation	1,335	~263	~1,072	19.7%
Test	1,336	~253	~1,083	18.9%
Total	13,354	~2,609	~10,745	19.5%

Label Mapping (Binary)

Original Class (dx)	Binary Label	Category
melanoma	1 (Malignant)	Malignant melanocytic
basal_cell_carcinoma	1 (Malignant)	Non-melanocytic malignant
actinic_keratoses	1 (Malignant)	Pre-cancerous
melanocytic_Nevi	0 (Benign)	Common moles
benign_keratosis-like_lesions	0 (Benign)	Seborrheic keratoses etc.
dermatofibroma	0 (Benign)	Benign fibrous
vascular_lesions	0 (Benign)	Angiomas etc.

MALIGNANT_CLASSES = {"melanoma", "basal_cell_carcinoma", "actinic_keratoses"}
label = 1 if item["dx"] in MALIGNANT_CLASSES else 0

Preprocessing

• Resize: 384x384 pixels
• Normalization: ImageNet stats (mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
• Augmentation (train only): RandomResizedCrop, Flips, Rotation, ColorJitter, CoarseDropout, CLAHE
• Oversampling: Malignant samples repeated 3x

Model Performance (DermaLens V3)

Metric	Value
Test ROC-AUC	0.9753
Test PR-AUC	0.9127
Test F1	0.8457
Sensitivity	94%
Specificity	91%

Model weights: dheraingoud/dermalens-model