Skin cancer detection using dermoscopic images with convolutional neural network

Area of Science:

• Dermatology and Medical Imaging
• Artificial Intelligence in Healthcare

Background:

• Malignant melanoma poses a high mortality risk, emphasizing the need for early detection.
• Existing machine learning methods for melanoma classification lack feature extraction depth, hindering accurate diagnosis.

Purpose of the Study:

• To introduce a deep learning network (FCDS-CNN) for enhanced skin lesion detection and data augmentation.
• To address class imbalance issues in melanoma datasets for improved diagnostic accuracy.

Main Methods:

• Developed a novel FCDS-CNN architecture incorporating data augmentation and class weighting.
• Utilized a dataset of 10,015 skin lesion images across seven classes from Kaggle.
• Implemented techniques to mitigate class imbalance and improve data quality.

Main Results:

• The FCDS-CNN achieved an average accuracy of 96%.
• Outperformed established models like ResNet, EfficientNet, Inception, and MobileNet in precision, recall, F1-score, and AUC.
• Demonstrated practical effectiveness in real-world application for early screening.

Conclusions:

• The FCDS-CNN offers a robust and scalable solution for early skin cancer detection.
• Highlights the importance of specialized deep learning models for nuanced medical image analysis.
• Supports dermatologists by providing a tool for improved early screening processes.