Semi-supervised skin cancer diagnosis based on self-feedback threshold focal learning

Area of Science:

• Dermatology
• Artificial Intelligence
• Medical Imaging

Background:

• Skin cancer diagnosis relies heavily on accurate image analysis.
• Supervised learning methods for AI-driven diagnosis require extensive labeled data, posing a significant limitation.
• Public health burdens associated with skin cancer necessitate more efficient diagnostic tools.

Purpose of the Study:

• To develop a semi-supervised learning model for skin cancer diagnosis that effectively utilizes both labeled and unlabeled medical images.
• To overcome the data labeling bottleneck inherent in traditional supervised learning approaches.
• To enhance the accuracy and efficiency of early skin cancer detection models, particularly in scenarios with limited annotated data.

Main Methods:

• Proposed a novel semi-supervised skin cancer diagnostic model named Self-feedback Threshold Focal Learning (STFL).
• Implemented dynamic adjustment of unlabeled sample selection thresholds to filter reliable data.
• Employed focal learning to address class imbalance issues during model training.
• Validated the model on the HAM10000 dataset across varying scales of labeled samples.

Main Results:

• The STFL model achieved robust performance with only 500 annotated samples, demonstrating 0.77 accuracy, 0.6408 Kappa, 0.77 recall, 0.7426 precision, and 0.7462 F1-score.
• Comprehensive testing confirmed significant advancements in diagnostic accuracy and efficiency by integrating unlabeled data.
• The model proved effective in unseen scenarios, highlighting its generalization capabilities.

Conclusions:

• Semi-supervised learning, particularly the STFL model, offers a powerful approach to enhance skin cancer diagnostic accuracy and efficiency.
• The integration of unlabeled data significantly mitigates the need for large labeled datasets, making AI diagnostics more accessible.
• This research provides a valuable tool and scientific support for the early diagnosis and treatment of skin cancer.