Intraclass Clustering-Based CNN Approach for Detection of Malignant Melanoma

Area of Science:

• Dermatology
• Oncology
• Artificial Intelligence
• Computer Vision

Background:

• Malignant melanoma is an aggressive skin cancer requiring early detection.
• Accurate classification of skin lesions is crucial for effective treatment.
• Convolutional Neural Networks (CNNs) show promise in medical image analysis.

Purpose of the Study:

• To develop and optimize a CNN model for detecting malignant melanoma.
• To improve the diagnostic accuracy of skin lesion classification.
• To achieve a high Receiver Operating Characteristic Area Under the Curve (ROC-AUC) score.

Main Methods:

• Fine-tuning a state-of-the-art CNN architecture.
• Utilizing artificial intelligence (AI) clustering techniques for model training.
• Training on a combined dataset from the 2019 and 2020 IIM-ISIC Melanoma Classification Challenges.
• Employing cross-fold validation for robust model evaluation.

Main Results:

• The optimized CNN model achieved a highest ROC-AUC score of 99.48%.
• The AI clustering techniques contributed to improved model performance.
• The model demonstrated high effectiveness in classifying melanoma from skin lesion images.

Conclusions:

• The developed CNN model offers a highly accurate method for malignant melanoma detection.
• AI-driven approaches, particularly CNNs, significantly advance dermatological diagnostics.
• This research provides a strong foundation for AI-assisted early skin cancer detection systems.