Skin lesion image segmentation based on lightweight multi-scale U-shaped network.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Deep learning models like UNet excel in medical image segmentation but demand substantial computational resources.
• Real-time applications necessitate a balance between segmentation accuracy and computational efficiency.

Purpose of the Study:

• To introduce a lightweight multi-scale U-shaped network (LMUNet) for efficient medical image segmentation.
• To address the trade-off between accuracy and computational complexity in medical image segmentation tasks, particularly for skin lesions.

Main Methods:

• The proposed LMUNet integrates multi-scale inverted residual blocks and an asymmetric atrous spatial pyramid pooling module.
• The network architecture is optimized for reduced parameter count and computational load.
• Performance was evaluated on various medical image segmentation datasets.

Main Results:

• LMUNet achieved a 67-fold reduction in parameters and a 48-fold decrease in computational complexity compared to existing methods.
• The proposed lightweight network demonstrated superior performance over other lightweight segmentation networks.
• Improved accuracy was observed in skin lesion image segmentation tasks.

Conclusions:

• LMUNet offers a computationally efficient solution for medical image segmentation, enabling real-time applications.
• The network effectively balances accuracy and computational complexity, outperforming existing lightweight alternatives.
• This approach is promising for advancing real-time medical image analysis, especially in dermatology.