Initial Geometrical Templates with Parameter Sets for Active Contour on Skin Cancer Boundary Segmentation.

Area of Science:

• Medical Imaging
• Computational Biology
• Dermatology

Background:

• Active contour models, or snake models, are crucial for skin cancer boundary segmentation.
• Segmentation success relies heavily on expert-initialized snake model starting points and algorithm parameters.
• Current methods are labor-intensive and require significant expert input.

Purpose of the Study:

• To propose automated initial geometrical templates and parameter sets for active contour-based skin cancer boundary segmentation.
• To reduce reliance on expert initialization and optimize algorithm parameters.
• To enhance the accuracy and efficiency of skin cancer segmentation.

Main Methods:

• Development of geometrically designed template candidates based on similarity to skin cancer boundaries.
• Selection of the best initial template by minimizing differences with the actual boundary.
• Determination of optimal parameter sets for each template through random variation and testing.
• Validation using experiments on 227 skin cancer samples.

Main Results:

• Achieved high segmentation performance: 99.46% accuracy, 97.43% sensitivity, and 99.87% specificity.
• Demonstrated improvements over conventional methods by 0.26% in accuracy, 0.36% in sensitivity, and 0.26% in specificity.
• Successfully automated the initialization and parameter setting process.

Conclusions:

• The proposed method effectively automates initial template and parameter set selection for active contour segmentation.
• This approach offers a significant improvement in accuracy, sensitivity, and specificity for skin cancer boundary segmentation.
• The findings suggest a more efficient and reliable tool for dermatological image analysis.