Skin Cancer Classification Using Convolutional Neural Networks: Systematic Review.

Area of Science:

• Dermatology
• Computer Science
• Medical Imaging

Background:

• Convolutional neural networks (CNNs) demonstrate dermatologist-level accuracy in skin cancer image classification.
• CNNs offer potential for rapid, accessible diagnoses via mobile applications.
• No prior reviews exist on the current state of CNNs for skin lesion classification.

Purpose of the Study:

• To systematically review the state-of-the-art research on classifying skin lesions using CNNs.
• To focus specifically on CNNs used for end-to-end skin lesion classification.
• To identify challenges in comparing methods and suggest future research directions.

Main Methods:

• Systematic literature search across major scientific databases (Google Scholar, PubMed, Medline, ScienceDirect, Web of Science).
• Inclusion of original research articles and systematic reviews published in English.
• Selection of papers reporting sufficient scientific proceedings for review.

Main Results:

• Thirteen papers were identified that classify skin lesions using CNNs.
• The most common and effective approach involves fine-tuning pre-trained CNNs on skin lesion datasets.
• This transfer learning method shows the best performance with limited available data.

Conclusions:

• CNNs are highly effective for skin lesion classification.
• Comparability of current methods is hindered by the use of non-public datasets, impacting reproducibility.
• Future studies should utilize public benchmarks and transparent methodologies for better comparison.