LDANet: Automatic lung parenchyma segmentation from CT images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate lung parenchyma segmentation in CT images is crucial for clinical diagnosis and treatment planning.
• Existing methods may face challenges in precisely delineating lung structures.

Purpose of the Study:

• To propose a novel deep learning algorithm, Lung Dense Attention Network (LDANet), for automatic lung parenchyma segmentation.
• To enhance segmentation accuracy by integrating spatial and channel attention mechanisms.

Main Methods:

• Development of LDANet incorporating Residual Spatial Attention (RSA) and Gated Channel Attention (GCA).
• Introduction of a Dual Attention Guidance Module (DAGM) for mechanism integration.
• Utilization of a Lightweight Dense Block (LDB) and Positioned Transpose Block (PTB) for feature reuse and resolution restoration.

Main Results:

• LDANet achieved high Dice similarity coefficients of 0.98430 on the LIDC-IDRI dataset and 0.98319 on the COVID-19 CT Segmentation dataset.
• The proposed model outperformed a state-of-the-art lung segmentation method.
• Ablation studies confirmed the effectiveness of LDANet's core components.

Conclusions:

• LDANet demonstrates superior performance in automatic lung parenchyma segmentation from CT images.
• The integration of RSA and GCA mechanisms significantly contributes to the model's accuracy.
• LDANet offers a promising tool for improving lung imaging analysis in clinical practice.