Automatic Lung Segmentation on Chest X-rays Using Self-Attention Deep Neural Network.

Area of Science:

• Medical Imaging
• Deep Learning
• Computer Vision

Background:

• Accurate segmentation of organs and abnormalities in medical images is crucial for surgical planning, diagnosis, and prognosis.
• Existing methods for medical image segmentation face challenges in precisely identifying boundaries, especially in complex structures like lungs.

Purpose of the Study:

• To propose a novel deep learning-based method for segmenting lung areas in chest X-ray images.
• To introduce and evaluate a self-attention module that combines channel and spatial attention for improved feature map highlighting.

Main Methods:

• A deep learning model incorporating a novel self-attention module was developed for lung segmentation.
• The self-attention module combines channel and spatial attention to generate attention maps, guiding the learning process.
• Attention maps are integrated into a U-Net architecture via residual learning, with experiments varying module placement within the network.

Main Results:

• The proposed method, when attention modules were placed in lower layers of the U-Net's contracting and expanding paths, achieved comparable or superior performance.
• Performance was evaluated using the Dice score on public chest X-ray datasets.
• The self-attention mechanism effectively highlighted relevant regions for improved segmentation accuracy.

Conclusions:

• The proposed deep learning method with self-attention modules demonstrates effective lung area segmentation in chest X-rays.
• Strategic placement of attention modules in lower network layers is key to achieving high performance, outperforming existing segmentation networks.
• This approach offers a promising advancement for automated medical image analysis and clinical applications.