Precision and efficiency in skin cancer segmentation through a dual encoder deep learning model.

Area of Science:

• Dermatology
• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Skin cancer is a significant global health issue.
• Accurate segmentation of skin lesions is vital for early diagnosis and effective treatment planning.
• Current segmentation methods often struggle to balance computational efficiency with comprehensive feature extraction.

Purpose of the Study:

• To introduce DuaSkinSeg, a novel deep learning model designed for improved skin lesion segmentation.
• To address the limitations of existing methods by enhancing both efficiency and feature extraction capabilities.
• To leverage a dual-encoder architecture for superior performance in segmenting skin lesions.

Main Methods:

• Developed DuaSkinSeg, a deep learning model featuring a dual-encoder architecture.
• Utilized a pre-trained MobileNetV2 for efficient local feature extraction.
• Employed a Vision Transformer-Convolutional Neural Network (ViT-CNN) encoder-decoder for high-level feature extraction and long-range dependency analysis.

Main Results:

• DuaSkinSeg demonstrated competitive performance across three benchmark datasets (ISIC 2016, 2017, 2018).
• The dual-encoder approach effectively combined local and global feature extraction for enhanced segmentation accuracy.
• Achieved comparable or superior results to existing state-of-the-art methods in skin lesion segmentation.

Conclusions:

• The proposed DuaSkinSeg model offers a promising solution for accurate and efficient skin lesion segmentation.
• The dual-encoder strategy represents a viable approach for improving deep learning models in medical image analysis.
• DuaSkinSeg has the potential to aid in the early and accurate diagnosis of skin cancer.