Outline of image analysis procedures. Step 1: A mask outlining the lung borders was obtained for each set of paired supine and prone images. Step 2: Each segmented prone mask was registered to the corresponding supine mask using, in sequence: rigid, affine, and deformable registration algorithms, and finally generating a transform (Φ_M). Step 3: Registration of the prone to supine image was performed, building on the previously obtained transform Φ_M. A map of the Jacobian was also created during this step to illustrate voxel-by-voxel lung volume changes due to position change. Step 4: The masks of the paired images were applied, and further image analysis was performed on the segmented target supine and warped-prone images (shown with inverted orientation to highlight similarities with the target supine image). Finally, subtraction maps and frequency distributions of intensity differences (ΔHU) were created for each pair of supine and prone images. Hounsfield Unit (HU) distributions were used to map lung recruitment and derecruitment with voxel resolution and, together with the Jacobian distributions, to perform cluster analysis of the inflation and deflation characteristics of each voxel.