Skin lesion image segmentation using Delaunay Triangulation for melanoma detection

Area of Science:

• Dermatology
• Medical Imaging
• Computer Vision

Background:

• Early detection of skin cancer, particularly melanoma, is crucial for reducing mortality.
• Automated diagnostic tools for skin lesions require accurate image segmentation.
• Current segmentation methods may have limitations in processing diverse lesion types.

Purpose of the Study:

• To develop a fast and fully-automatic algorithm for skin lesion segmentation in dermoscopic images.
• To evaluate the algorithm's performance against state-of-the-art methods.
• To explore the use of segmentation-derived features for melanoma classification.

Main Methods:

• Implementation of a novel segmentation algorithm utilizing Delaunay Triangulation.
• Extraction of a binary mask for skin lesion regions without a training stage.
• Quantitative experimental evaluation on a public dermoscopic image database.

Main Results:

• The proposed algorithm demonstrated high accuracy in segmenting benign skin lesions.
• Segmentation accuracy significantly decreased when processing melanoma images.
• Geometrical and color features extracted from masks showed promise for melanoma classification.

Conclusions:

• The Delaunay Triangulation-based segmentation is effective for benign lesions but requires enhancement for melanoma.
• Feature extraction from segmentation masks offers a viable strategy for improving melanoma detection.
• Further development is needed to address the challenges in segmenting and classifying melanoma images.