Advancing Skin Cancer Detection Integrating a Novel Unsupervised Classification and Enhanced Imaging Techniques

Paper

📄 This code implements the paper published in a Q1 Journal, CAAI Transactions on Intelligence Technology, with an impact factor of 8.4.

Title: Advancing skin cancer detection integrating a novel unsupervised classification and enhanced imaging techniques

Methodology

Our methodology is a combination of modified ESRGAN, a novel histogram feature extraction map, optimal cluster-number estimation, and the application unsupervised clustering algorithm.

Dataset

The following two public datasets were used in our experiment:

• ISIC 2019: andrewmvd/isic‐2019
• HAM10000: kmader/skin-cancer-mnist-ham10000

ESRGAN

The detailed implentation of ESR‐Generative Adversarial Networks is available at Code 📁, and experimented other pretrained model can be found Here 📁.

Histogram Feature Map

Histogram feature map generation and extraction details is available at Code 📁

Optimal Cluster Number

Experimented optimal number of cluster finding with DBI and SS score evaluation code is detailed at Code 📁

Clustering Performance Evaluation

k-Means clusteing algorithm was finally choosen for our appraoch after exploring several other clusting algorithm. Code is avalable at Code 📁

Requirements

The following key Python packages are required to run the code:

• TensorFlow
• PyTorch
• NumPy
• Keras
• Pandas
• Matplotlib

Citation Request

If you find this work helpful for your research, please consider citing our paper:

• Cite:

@article{rahman2025advancing,
  title = {Advancing skin cancer detection integrating a novel unsupervised classification and enhanced imaging techniques},
  author = {Rahman, Md. Abdur and Fahad, Nur Mohammad and Raiaan, Mohaimenul Azam Khan and Jonkman, Mirjam and De Boer, Friso and Azam, Sami},
  journal = {CAAI Transactions on Intelligence Technology},
  pages = {1--20},
  year = {2025},
  doi = {10.1049/cit2.1241020},
  publisher = {Wiley}
}