Multimodal skin lesion classification for early cancer diagnosis using deep learning.

Area of Science:

• Dermatology and Artificial Intelligence
• Medical Image Analysis
• Computational Pathology

Background:

• Skin cancer, especially melanoma, is a significant global health concern.
• Early detection of skin cancer dramatically improves survival rates and treatment efficacy.
• Traditional diagnostic methods can be resource-intensive and rely on expert interpretation.

Purpose of the Study:

• To develop and evaluate a deep learning model for accurate and efficient skin cancer detection.
• To leverage Deep Convolutional Neural Networks (DCNNs) for classifying skin lesions from dermoscopic images.
• To enhance diagnostic accuracy and assist dermatologists in early skin cancer identification.

Main Methods:

• Utilized the HAM10000 dataset comprising dermoscopic images of skin lesions.
• Employed three DCNN architectures: DenseNet-201, VGG16, and InceptionV3.
• Implemented preprocessing, fine-tuning strategies, model ensembling, and Grad-CAM for interpretability.

Main Results:

• The ensemble model achieved a testing accuracy of 97.9%.
• The model demonstrated superior performance with an F1-score, recall, and precision of 99.2%.
• The ensemble approach outperformed individual DCNN models in skin lesion classification.

Conclusions:

• Ensemble deep learning models show high efficacy in automated skin lesion detection.
• The developed model offers a promising tool for early and accurate skin cancer diagnosis.
• Enhanced model interpretability through Grad-CAM increases clinical applicability and trust.