An improved V-Net lung nodule segmentation model based on pixel threshold separation and attention mechanism

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate lung nodule segmentation in computed tomography (CT) is vital for early lung cancer diagnosis and surgical planning.
• The irregular shapes of nodules and complex lung environment pose significant segmentation challenges.

Purpose of the Study:

• To propose an improved V-Net segmentation method for enhanced lung nodule detection in CT images.
• To address challenges in 3D medical dataset sample size and improve segmentation accuracy.

Main Methods:

• Developed an improved V-Net model incorporating pixel threshold separation and attention mechanisms.
• Implemented a data augmentation strategy to overcome limited 3D medical dataset samples.
• Integrated feature extraction modules and channel/spatial attention to boost model performance.

Main Results:

• Achieved Dice similarity coefficients of 94.9% (LUNA16) and 81.1% (LNDb).
• Reached sensitivities of 92.7% (LUNA16) and 76.9% (LNDb).
• Demonstrated superior performance compared to existing UNet models and comparable results to manual segmentation.

Conclusions:

• The proposed V-Net model effectively improves lung nodule segmentation accuracy and generalization.
• This method shows significant potential for enhancing early lung cancer diagnosis through precise nodule identification.