Analysis of segmentation of lung parenchyma based on deep learning methods

Area of Science:

• Medical Imaging
• Computer Vision
• Radiology

Background:

• Accurate lung parenchyma segmentation is critical for effective lung analysis using medical imaging.
• Computed tomography (CT) and computed tomography angiography (CTA) are standard modalities for lung imaging.
• Existing segmentation methods require improvement in speed and quality.

Purpose of the Study:

• To evaluate and analyze lung parenchyma segmentation methods presented at the 4th International Symposium on Image Computing and Digital Medicine (ISICDM 2020).
• To identify the most effective deep learning approaches for segmenting lung parenchyma in CT and CTA images.

Main Methods:

• Twelve research teams participated, employing various segmentation techniques.
• The U-Net architecture or its modified versions were utilized by nine teams.
• Methods like attention mechanisms and multi-scale feature fusion were explored to enhance accuracy.

Main Results:

• The U-Net network achieved the highest performance, with a Dice coefficient of 0.991 for CT and 0.984 for CTA.
• Attention U-Net and nnU-Net also demonstrated strong segmentation results.
• The study provides an evaluation of different teams' methods and their segmentation outcomes.

Conclusions:

• Deep learning models, particularly U-Net and its variants, are highly effective for lung parenchyma segmentation.
• The findings offer valuable insights and references for researchers and clinicians in medical image analysis.
• Optimized segmentation techniques contribute to improved diagnostic capabilities in pulmonary imaging.