Skin Cancer Detection with CNN

Project Summary:

The project aims to develop a CNN-based tool to predict the probability of skin moles being malignant. Skin cancer is a prevalent disease, and early detection is crucial for improving survival rates. The tool will analyze images of moles and provide a probability score for malignancy.

Development Process and Data:

Data Source:

Utilizing datasets from the International Skin Imaging Collaboration: Melanoma Project (ISIC).

Datasets Used:

Various datasets containing biopsy-confirmed melanocytic lesions, including both malignant and benign cases.

Preprocessing:

Includes visual inspection for quality, resizing images to 128x128x3, and cropping for better representation.

CNN Model:

Developing a CNN model from scratch and enhancing it through techniques like data augmentation and transfer learning (using pre-trained networks like VGG-16).

Model Evaluation:

Utilizing ROC curves and AUC scores for model evaluation, with a focus on precision, accuracy, and finding the optimal threshold for tradeoff between true positive and false positive rates.

Motivation:

Skin cancer affects a large population, with alarming statistics highlighting the need for effective detection tools. By leveraging CNN technology, this project aims to contribute to early detection efforts, potentially saving lives and improving patient outcomes.

Developer:

Pavan Kumar CH([email protected])

Program Affiliation:

360digitmg Professional Data Science & AI Course.

Data Source:

International Skin Imaging Collaboration: Melanoma Project (ISIC) - Link to ISIC Archive

Sample Images:

Sample images of benign and malignant moles

Next Steps:

Continued refinement and optimization of the CNN model, integration into a user-friendly tool for medical professionals, and potential deployment for wider public use.

Note:

All information and data provided in this project summary are for educational and informational purposes only. The project does not provide medical advice or diagnosis.