Development and Validation of Adaptable Skin Cancer Classification System Using Dynamically Expandable...

Area of Science:

• Dermatology
• Artificial Intelligence
• Machine Learning

Background:

• Skin cancer is a common malignancy requiring efficient diagnostic methods.
• Automated classification systems can aid in early and accurate diagnosis.

Purpose of the Study:

• To develop an automated skin lesion classification model using the dynamically expandable representation (DER) incremental learning algorithm.
• To create a scalable and efficient system for skin cancer diagnosis.

Main Methods:

• The DER incremental learning model was applied to the HAM10000 and ISIC 2019 datasets.
• Model validation included internal training/evaluation on HAM10000 and external validation on ISIC 2019.
• Performance metrics included precision, recall, F1-score, and area under the precision-recall curve (AUC).

Main Results:

• The model achieved a weighted-average precision of 0.918, recall of 0.808, and F1-score of 0.847.
• The average area under the curve (AUC) was 0.943, indicating strong discrimination performance.
• External validation on the ISIC 2019 dataset yielded an AUC of 0.911, confirming effectiveness.

Conclusions:

• The DER algorithm-based skin lesion classification model demonstrates high performance and scalability.
• The model shows potential for expanding its classification range and adapting to new disease classes.
• Robust external validation results support the model's clinical applicability in skin cancer diagnosis.