Early Melanoma Detection Based on a Hybrid YOLOv5 and ResNet Technique

Area of Science:

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

Background:

• Melanoma, a form of skin cancer, poses a significant health risk due to rising incidence rates.
• Accurate and timely identification of skin lesions is critical for effective melanoma treatment.
• Computer-assisted diagnostic tools are being developed to address challenges in classifying similar-looking benign and malignant skin lesions.

Purpose of the Study:

• To develop and evaluate a novel computer-assisted diagnostic model for early melanoma detection.
• To improve the accuracy and efficiency of classifying dermatoscopic images of skin lesions.
• To create a smaller, more precise deep learning model for melanoma identification.

Main Methods:

• A hybrid deep learning model integrating YOLOv5 (You Only Look Once) and ResNet50 architectures was developed.
• Feature maps were enhanced with gradient change for rapid inference and reduced hyperparameters.
• The YOLOv5 model was adapted with new classes for pigmented skin lesions using the HAM10000 dataset.

Main Results:

• The proposed model achieved high performance metrics, including 99.0% precision, 98.6% recall, and 99.5% accuracy.
• Real-time detection speed was achieved at 0.4 milliseconds per image (non-maximum suppression).
• The model demonstrated superior efficiency in utilizing deep features compared to existing melanoma detection methods.

Conclusions:

• The combined YOLOv5 and ResNet50 model offers a highly accurate and efficient solution for melanoma detection.
• This AI-driven approach shows promise in assisting clinicians with early and reliable skin cancer diagnosis.
• The model's improved precision and speed contribute to advancing computer-assisted diagnostics in dermatology.