Integrating Patient Data Into Skin Cancer Classification Using Convolutional Neural Networks: Systematic Review

Area of Science:

• Medical Imaging
• Artificial Intelligence in Dermatology
• Computational Pathology

Background:

• Convolutional Neural Networks (CNNs) show high accuracy in single-image skin cancer classification, matching or exceeding dermatologists.
• Clinical diagnosis integrates patient data beyond images, enhancing dermatologist accuracy.
• Pilot studies explore integrating diverse patient data into CNN-based skin cancer classifiers.

Purpose of the Study:

• Systematically review research on merging image features and patient data for CNN-based skin cancer classification.
• Evaluate patient data types, encoding/merging methods, and their impact on classifier performance.
• Explore the potential of integrated data approaches in advancing skin cancer diagnostics.

Main Methods:

• Systematic literature search across Google Scholar, PubMed, MEDLINE, and ScienceDirect.
• Included peer-reviewed English studies on integrating patient data into CNN skin cancer classification.
• Combined search terms: skin cancer classification, CNNs, deep learning, lesions, melanoma, metadata, clinical information, patient data.

Main Results:

• 11 publications met inclusion criteria, all reporting performance improvements with patient data integration.
• Commonly used patient data included age, sex, and lesion location; typically one-hot encoded.
• Varied complexity in processing encoded patient data before and after fusion with image features.

Conclusions:

• Integrating patient data into CNN diagnostic algorithms offers significant potential benefits.
• The precise mechanisms by which patient data enhance classification, especially for multiclass problems, require further investigation.
• Substantial patient data utilized by dermatologists remain unanalyzed in CNN contexts; further research can optimize integration for patient benefit.