LLNS-Net: Lightweight Lung Nodule Segmentation Network With Multiscale Information Fusion and Complementarity

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Accurate lung nodule segmentation is crucial for early lung cancer detection.
• Existing algorithms struggle to balance high accuracy with lightweight design for clinical application.

Purpose of the Study:

• To propose LLNS-Net, a compact and effective network for lung nodule segmentation.
• To improve the accuracy and efficiency of lung nodule segmentation in CT images.

Main Methods:

• Developed LLNS-Net featuring a feature-mining encoder with multiscale attention and a feature enhancement module.
• Incorporated an enhanced mixed local channel attention (E-MLCA) mechanism and a reinforced multiscale feature module.
• Utilized a decoder with subchannel enhancement for refining segmentation maps and improving boundary smoothness.

Main Results:

• LLNS-Net achieved improved intersection over union (IoU) compared to HmsUnet (1.86%), MSA-Unet (0.27%), and H-vmunet (7.62%).
• The network generated feature maps with smoother boundaries and superior visual quality.
• Demonstrated a balance between segmentation accuracy and network compactness.

Conclusions:

• LLNS-Net presents a promising solution for efficient and accurate lung nodule segmentation.
• The proposed architecture enhances feature learning and segmentation refinement for better morphological preservation.
• LLNS-Net offers a competitive alternative to existing methods in medical image segmentation.