Unified approach for lesion border detection based on mixture modeling and local entropy thresholding

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Dermatology

Background:

• Dermatologist diagnostic accuracy for skin lesions is often low.
• Digital dermoscopy and computer-aided design (CAD) methods offer valuable analysis tools.
• Automatic border detection (ABD) is a critical initial step in computerized lesion analysis.

Purpose of the Study:

• To propose a novel unified approach for automatic border detection (ABD) in digital dermoscopy.
• To enhance the accuracy of skin lesion segmentation in computerized diagnostic methods.

Main Methods:

• A normalized smoothing filter (NSF) was used for preprocessing to reduce noise.
• Mixture models technique was employed for initial rough segmentation of the lesion area.
• Local entropy thresholding and morphological reconstruction were utilized for precise border extraction and smoothing.

Main Results:

• The proposed ABD system was evaluated on 100 dermoscopy images with ground truth.
• Comparative analysis against three state-of-the-art methods demonstrated superior performance.
• The technique achieved a minimal average error probability of 5%, 92.10% true detection, and a 6.41% false positive rate.

Conclusions:

• The developed method accurately segments skin lesion areas.
• The findings suggest a significant improvement in automated skin lesion analysis.
• Sample datasets and software are available online for further research and application.