Deep learning segmentation and registration-driven lung parenchymal volume and movement CT analysis in prone...

Area of Science:

• Radiology and Imaging
• Pulmonary Medicine
• Artificial Intelligence in Medicine

Background:

• Prone positioning is used to improve oxygenation in patients with acute respiratory distress syndrome.
• Understanding lung parenchyma changes during prone positioning is crucial for optimizing patient management.
• Deep neural networks (DNNs) offer advanced capabilities for medical image analysis.

Purpose of the Study:

• To analyze volumetric and movement changes in lung parenchyma during prone positioning using DNNs.
• To quantify lung lobe expansion and pixel movement between supine and prone positions.
• To investigate the impact of lung fibrosis on these positional changes.

Main Methods:

• Chest CT scans from 108 patients with suspected interstitial lung disease were analyzed.
• A DNN automatically segmented lung lobes in supine and prone positions.
• Affine and deformable registration calculated lung volume changes and pixel movements.

Main Results:

• Prone positioning significantly increased left and right lower lobar volumes.
• Average maximum whole-lung pixel movements were calculated across axes and in 3D.
• Patients with ≥30% lung fibrosis exhibited smaller volume changes and pixel movements.
• Forced vital capacity (FVC) correlated with lower lobar volume increase and lung movements.

Conclusions:

• Prone positioning preferentially expands lower lung lobes.
• Lung fibrosis limits lung expansion during prone positioning.
• FVC is associated with the degree of lung expansion and movement in prone position.