Automatic skin lesion segmentation by coupling deep fully convolutional networks and shallow network with textons.

Area of Science:

• Medical image analysis
• Computer-aided diagnosis
• Dermatology

Background:

• Skin lesion segmentation is crucial for melanoma diagnosis but challenging due to fuzzy boundaries and textures.
• Existing methods often struggle with the complexity of skin lesion characteristics.

Purpose of the Study:

• To develop a fully automatic method for accurate skin lesion segmentation.
• To enhance computer-aided diagnosis of melanoma by improving segmentation accuracy.

Main Methods:

• A deep fully convolutional network (FCN) approach was employed.
• A shallow encoding network was used to incorporate clinically valuable prior knowledge.
• A fusing strategy combined shallow network prior knowledge with deep FCN features using skip connections.

Main Results:

• The method demonstrated improved accuracy in skin lesion segmentation on ISBI 2016 and 2017 datasets.
• The integration of prior knowledge with deep FCNs enhanced segmentation performance.
• The approach showed robustness, effective model generalization, and outperformed state-of-the-art methods.

Conclusions:

• The proposed method effectively segments skin lesions by coupling prior knowledge with deep FCNs.
• This approach offers a promising advancement for computer-aided melanoma diagnosis.
• The method is robust and generalizes well without extensive parameter tuning or data augmentation.