Skin Cancer Classification Using CNN

A deep learning solution for detecting nine types of skin cancer, including melanoma, using a custom Convolutional Neural Network (CNN). This project aims to improve early detection of melanoma and other skin cancers by automating the classification of dermoscopic images.

Table of Contents

• General Info
• Technologies Used
• Conclusions
• Acknowledgements

General Information

• Project Background:

  Skin cancer is one of the most common forms of cancer, and melanoma accounts for 75% of skin cancer deaths. Early detection is critical to improving survival rates. However, manual diagnosis by dermatologists is time-consuming and prone to human error. This project provides an automated solution using a custom CNN to classify skin cancer types based on dermoscopic images.

• Business Problem:

  The goal is to build a machine learning model that can evaluate skin lesion images and accurately classify them into one of nine categories, reducing the workload on dermatologists and enabling quicker diagnosis.

• Dataset Information:

  The dataset used in this project contains 2,357 images of nine different skin cancer types. It was sourced from the International Skin Imaging Collaboration (ISIC). The dataset includes:

  • Actinic keratosis
  • Basal cell carcinoma
  • Dermatofibroma
  • Melanoma
  • Nevus
  • Pigmented benign keratosis
  • Seborrheic keratosis
  • Squamous cell carcinoma
  • Vascular lesion

  Images were augmented to balance class distribution and improve model generalization.

Conclusions

1. Class Rebalancing Improved Performance:

Augmenting underrepresented classes using the Augmentor library reduced overfitting and improved validation accuracy slightly.

2. Challenges in Model Complexity:

The current CNN architecture shows signs of underfitting, indicating the need for a more sophisticated model to capture complex features.

3. Generalization Remains a Challenge:

Validation accuracy reached ~49%, which suggests further improvements are needed in both data preprocessing and model design.

4. Potential for Practical Use:

The automated classification system demonstrates the feasibility of using CNNs for dermatological applications, though further refinement is required to achieve clinically relevant accuracy levels.

Technologies Used

• Python - version 3.10
• TensorFlow - version 2.12
• Augmentor - version 0.2.10
• Matplotlib - version 3.7
• NumPy - version 1.23
• Pandas - version 1.4

Acknowledgements

Give credit here.

• This project was inspired by the need to enhance medical diagnostics with AI.
• Dataset: International Skin Imaging Collaboration (ISIC).
• Based on tutorials from TensorFlow documentation and Augmentor library documentation.

Contact

Created by [@ManisCodeBase] - feel free to contact me!