An automatic method for lung segmentation and reconstruction in chest X-ray using deep neural networks.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Artificial Intelligence in Medicine

Background:

• Chest X-rays (CXRs) are crucial for diagnosing pulmonary diseases.
• Automatic lung segmentation in CXRs is vital for computer-aided detection and diagnosis.
• Dense opacities obscure lung regions, challenging accurate segmentation.

Purpose of the Study:

• To propose a novel method for automatic lung segmentation in CXRs.
• To address the challenge of segmenting lung regions obscured by dense abnormalities.
• To reconstruct 'lost' lung regions due to pulmonary abnormalities.

Main Methods:

• The method employs two deep convolutional neural network models.
• It involves four main steps: image acquisition, initial segmentation, reconstruction, and final segmentation.
• A reconstruction step is key to addressing obscured lung regions.

Main Results:

• The method achieved high performance on 138 CXR images.
• Results include average sensitivity of 97.54%, specificity of 96.79%, and accuracy of 96.97%.
• Excellent segmentation metrics were obtained: 94% Dice coefficient and 88.07% Jaccard index.

Conclusions:

• Dense abnormalities in CXRs can be effectively managed using this lung segmentation method.
• A reconstruction step, powered by deep convolutional neural networks, is key to the method's success.
• The proposed approach enhances the accuracy of lung segmentation in the presence of pulmonary abnormalities.