Skin Cancer Image Classification Using Artificial Intelligence Strategies: A Systematic Review

Area of Science:

• Dermatology
• Medical Imaging
• Artificial Intelligence

Background:

• Malignant skin tumors pose a significant public health challenge, with rising incidence and mortality.
• Current diagnostic methods rely heavily on physician expertise, risking delayed detection or unnecessary procedures.

Purpose of the Study:

• To systematically review the implementation of artificial intelligence (AI) methods for skin tumor classification using various imaging modalities.
• To assess the diagnostic potential and effectiveness of AI in improving skin cancer detection.

Main Methods:

• A systematic review following PRISMA guidelines was conducted.
• 206 records were retrieved and qualitatively analyzed, focusing on AI applications in skin tumor classification.
• Analysis included various imaging modalities like confocal and multiphoton laser scanning microscopy, and dermoscopy.

Main Results:

• AI, especially when combined with dermoscopy images, demonstrated significant diagnostic potential, achieving near-perfect classification accuracy.
• Support vector machines and artificial neural networks are commonly used learning approaches, with a recent trend towards convolutional neural networks.
• Some studies lacked detailed reporting of training/testing conditions, hindering reproducibility.

Conclusions:

• AI methods show great promise for improving the reliability and speed of skin cancer diagnosis.
• Further research is needed to optimize AI learning approaches and ensure reproducibility.
• Enhanced AI-driven early detection can significantly improve patient outcomes, even in resource-limited healthcare settings.