BiDFDC-Net: a dense connection network based on bi-directional feedback for skin image segmentation

Area of Science:

• Medical Imaging
• Computer Vision
• Dermatology

Background:

• Accurate skin lesion segmentation is crucial for patient survival rates.
• Challenges include blurred boundaries, diverse features, and cellular mutations in skin images.
• Existing segmentation algorithms lack effectiveness and robustness.

Purpose of the Study:

• To propose a robust and accurate skin lesion segmentation framework.
• To address challenges of gradient vanishing and information loss in deep networks.
• To enhance feature propagation and multi-level context information fusion.

Main Methods:

• Developed a bi-directional feedback dense connection network (BiDFDC-Net) based on the U-Net architecture.
• Integrated edge modules into the encoder to mitigate gradient vanishing and information loss.
• Employed dense connections for enhanced feature propagation and reuse.
• Utilized a two-branch decoder for multi-scale feature and context information fusion.

Main Results:

• Achieved 93.51% accuracy on the ISIC-2018 dataset.
• Achieved 94.58% accuracy on the PH2 dataset.
• Demonstrated improved effectiveness and robustness in skin lesion segmentation.

Conclusions:

• BiDFDC-Net effectively segments skin lesions in dermoscopic images.
• The proposed architecture overcomes limitations of existing methods.
• This approach holds significant potential for improving diagnostic accuracy in dermatology.