LightAWNet: Lightweight adaptive weighting network based on dynamic convolutions for medical image segmentation

Area of Science:

• Medical Image Analysis
• Computer Vision
• Deep Learning

Background:

• Convolutional Neural Networks (CNNs) offer high accuracy in medical image segmentation but suffer from complexity and resource demands.
• Lightweight models provide efficiency but often compromise segmentation accuracy.
• Balancing efficiency and accuracy is crucial for practical medical image analysis.

Purpose of the Study:

• To propose LightAWNet, a lightweight adaptive weighting neural network for medical image segmentation.
• To address the trade-off between model complexity and segmentation performance.
• To develop a resource-efficient yet highly accurate segmentation model.

Main Methods:

• Designed an efficient inverted bottleneck encoder block with spatial attention optimization.
• Employed a two-branch strategy for separate detailed and spatial feature extraction and fusion.
• Utilized a lightweight optimized up-sampling operation and channel attention in the decoder.

Main Results:

• LightAWNet achieved state-of-the-art performance on multiple datasets (LiTS2017, MM-WHS, ISIC2018, Kvasir-SEG) with only 2.83 million parameters.
• Demonstrated significantly improved segmentation accuracy compared to existing methods.
• Highlighted effectiveness in maintaining high performance with reduced model complexity.

Conclusions:

• LightAWNet successfully balances efficiency and accuracy in medical image segmentation.
• Innovative components like spatial attention and optimized up-sampling contribute to superior performance.
• Provides valuable insights for developing efficient and accurate medical imaging segmentation models.