Hybrid Deep Learning Framework for Melanoma Diagnosis Using Dermoscopic Medical Images.

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Oncology

Background:

• Melanoma, a dangerous form of skin cancer, is a significant global health concern.
• Early and accurate diagnosis is crucial for improving patient outcomes.
• Deep learning methods are increasingly utilized for medical image analysis.

Purpose of the Study:

• To propose a hybrid deep learning approach for early and accurate diagnosis and classification of skin cancer.
• To leverage advanced deep learning architectures for enhanced melanoma detection.

Main Methods:

• A hybrid deep learning model combining U-Net for segmentation, Inception-ResNet-v2 for feature extraction, and Vision Transformer with self-attention for feature refinement.
• Hyperparameter tuning was employed to optimize classification accuracy.
• The model was trained and validated on the ISIC2020 and HAM10000 datasets.

Main Results:

• The proposed approach achieved high performance metrics: 98.65% accuracy, 99.20% sensitivity, and 98.03% specificity on the ISIC2020 dataset.
• Performance surpassed existing methods for skin cancer classification.
• Ablation studies on the HAM10000 dataset validated the model's effectiveness.

Conclusions:

• The developed hybrid deep learning model demonstrates significant potential for the early detection of melanoma.
• This approach can serve as a valuable tool to assist dermatologists in clinical practice.
• The findings highlight the efficacy of integrating multiple deep learning architectures for complex medical diagnoses.