LSCS-Net: A lightweight skin cancer segmentation network with densely connected multi-rate atrous convolution

Area of Science:

• Medical Image Analysis
• Computer Vision
• Dermatology

Background:

• Skin cancer, particularly melanoma, presents significant public health challenges.
• Existing computer vision models face difficulties segmenting diverse dermatological images due to variations in lighting, patient characteristics, and hair density.

Purpose of the Study:

• To develop an innovative, end-to-end trainable network for enhanced skin cancer segmentation.
• To address the limitations of current methods in handling diverse and complex dermatological image data.

Main Methods:

• An encoder-decoder architecture was employed, incorporating a novel feature extraction block and a densely connected multi-rate Atrous convolution block.
• The proposed lightweight skin cancer segmentation network (LSCS-Net) was evaluated on ISIC 2016, ISIC 2017, and ISIC 2018 benchmark datasets.

Main Results:

• LSCS-Net achieved state-of-the-art results on skin lesion segmentation benchmarks.
• The network demonstrated excellent generalization capabilities on breast cancer and thyroid nodule segmentation datasets.
• A significantly elevated Jaccard index was observed, confirming the model's superior performance.

Conclusions:

• The developed LSCS-Net offers an advanced approach for accurate skin cancer segmentation.
• This AI-driven tool has the potential to improve early detection and management of skin cancer.
• The network's robust performance across multiple datasets highlights its clinical applicability.