Enhanced U-Net with Attention Mechanisms for Improved Feature Representation in Lung Nodule Segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Computer Vision

Background:

• Accurate segmentation of pulmonary nodules is crucial for lung cancer diagnosis.
• Traditional U-Net models face challenges with small, irregular nodules and complex backgrounds.
• Limitations include difficulty capturing long-range dependencies and integrating multi-scale features.

Purpose of the Study:

• To propose an enhanced U-Net hybrid model for precise pulmonary nodule segmentation.
• To improve feature representation and segmentation accuracy using integrated attention mechanisms.
• To address the limitations of standard U-Net models in challenging lung imaging scenarios.

Main Methods:

• Utilized the LUNA16 dataset of annotated CT scans for model assessment.
• Integrated multiple attention mechanisms: Spatial Attention (SA), Dilated Efficient Channel Attention (Dilated ECA), CBAM, and SE Block into a U-Net backbone.
• Designed architecture and training to specifically target small and irregular pulmonary nodule segmentation.

Main Results:

• Achieved a Dice similarity coefficient of 84.30%, demonstrating superior performance.
• Significantly outperformed the baseline U-Net model in nodule segmentation accuracy.
• Showcased improved precision for small and irregular pulmonary nodules.

Conclusions:

• The enhanced U-Net hybrid model effectively captures local and global features via integrated attention mechanisms.
• SA, Dilated ECA, CBAM, and SE modules collectively enhance segmentation performance in complex backgrounds.
• Future research may explore performance in cases with extreme anatomical variability or low-contrast lesions.