Automatic Skin Cancer Detection Using Clinical Images: A Comprehensive Review

Area of Science:

• Dermatology
• Computer Science
• Medical Imaging

Background:

• Skin cancer, particularly melanoma, is a growing concern, necessitating early detection.
• Machine learning (ML) shows promise for skin cancer identification, but most research uses dermoscopy images.
• General practitioners often lack dermoscopes, relying on standard clinical images for diagnosis.

Purpose of the Study:

• To comprehensively review image-processing techniques for skin cancer detection using clinical images.
• To evaluate 51 recent articles focusing on ML methods for skin cancer detection in clinical datasets.

Main Methods:

• Systematic review and analysis of 51 state-of-the-art research articles.
• Focus on studies utilizing machine learning for skin cancer detection from clinical images.

Main Results:

• Few publicly available clinical image datasets exist for benchmarking compared to dermoscopy datasets.
• Current artifact removal techniques in ML models can be inadequate and negatively impact performance.
• Most studies analyze single-lesion images, neglecting patient-specific mole patterns and temporal changes.

Conclusions:

• There is a significant need for larger, standardized clinical image datasets for robust ML model development.
• Future research should address artifact removal challenges and incorporate longitudinal data for improved skin cancer detection accuracy.