Mimics version 21

The images of the upper airway structure within the range from the cranial crest to the clavicle were obtained by CBCT scanner (KaVo 3D eXam, USA): a single 360° rotation scan, 120 kV voltage, 5 mA current, slice thickness 0.3 mm, scanning time 17.8 s. The midpoint cross line of the interpupil line overlapped with the central cross line of the location line during scanning. All of the images were imported into Mimics version 21.0 (Materialise Inc., Belgium. www.materialise.com) and then upper airway three-dimensional models were rebuilt. The volume, cross-sectional areas, sagittal diameter and cross diameter from the top level of the soft palate to the level of 1/4, 2/4 and 3/4 in the posterior airway were measured. PSG monitoring was performed as previously described^{12 (link)} (Fig. S1).

Chest CT scans were completed within 48 hours after admission for each participant using a 16-slice spiral CT scanner (Brilliance; Philips Healthcare, OH, USA) with a 5-mm slice thickness. Acquisition parameters were as follows: 100-140 kV, variable mA based on the patient’s body size, and detector collimation of 0.75-1.5 mm. Unenhanced cross-sectional CT images at the T12 level were analyzed using a dedicated segmentation software (Mimics version 21.0; Materialise, Leuven, Belgium) to evaluate the T12 SMA.
On a single CT image, all visualized skeletal muscles with a threshold of −29 to +150 HU, including the erector spinae, latissimus dorsi, rectus abdominis, obliquus externus, internus abdominis, and internal and external intercostal muscles, were segmented (12 (link)). The T12 SMI (cm²/m²) was calculated as the T12 SMA (cm²) divided by the body height squared (m²).
A trained observer (L.T.) who was blinded to patient outcomes during the analysis period segmented all CT images. To test the reliability of the T12 SMA determined by CT, a total of 30 participants were randomly selected from the cohort. To assess inter-observer reproducibility, another trained observer (S.H.) subsequently segmented the CT images again. Representative images are presented in eFigure 1.

The patient’s CTA data were imported into Materialise Mimics version 21.0 (Leuven, Belgium) software, and the threshold segmentation function was used to segment the 3D reconstruction. Then, the obtained 3D reconstructive model was extracted, trimmed, smoothed, and repaired digitally in the Materialise 3-Matic (Leuven, Belgium) software. The structure of the aortic root was restored completely. The standard tessellation language (STL) files of the 3D reconstruction were exported to a Stratasys Polyjet 850 multimaterial full-color 3D printer. Different tissues of the aortic root were printed with materials of different hardnesses and colors to obtain the patient’s 3D-printed aortic root model (Figure 1).