This cross-sectional study compared obese adolescents with normal pulmonary function, obese adults with normal pulmonary function, and obese adults with ventilation disorders.
We enrolled obese participants(BMI >30 kg/m2) 14 to 57 years of age with the following exclusion: (1) history of thoracic surgery and lung disease, (2) current or former smokers, (3) history of thoracic congenital skeletal abnormality, (4) occupational history that may lead to impairment of lung function. The study was approved by the institutional review board of the First Afﬁliated Hospital of Jinan University and informed written consent was obtained from all participants.
CT Image Acquisition
Unenhanced chest computed tomography (CT) was acquired using a 320-row CT scanner (Toshiba Aquilion ONE, Japan) from the apex to the lung base in the supine position during a brief breath-hold after inspiration. All participants received respiratory training prior to the CT examination to ensure successful breathing maneuvers during image acquisition. CT parameters for image acquisition were as follows:140 peak kVp, variable mAs, 0.5-second gantry rotation, the pitch of 1.0 or 1.25mm slice thickness, contiguous slices. Reconstruction was performed with a high spatial frequency algorithm (FC56).
CT Image Analysis
Airway Segmentation via Deep Learning
Based on the work of Minghui Zhang et al. , we first preprocess the CT scans and train an encoder-decoder network (), cooperating with the distance-related loss function to segment the airway tree structure, aiming to preserve the completeness of the airway tree as much as possible. Specifically, during the preprocessing, we set the window level at -300 HU with a width of 900 HU and rescale voxel values into [0, 255]. Besides, we cropped the lung field to remove unrelated background regions. The encoder extracts the semantic feature of the input image, and the symmetrical decoder receives the output of the encoder and the skip connection to predict the category of each voxel (i.e., airway or background). Secondly, we conducted a manual edition by two experienced radiologists based on the previous automatic airway segmentation step. The overall framework can be seen in Fig 1.
Fig 2 shows the segmentation results by different methods. a) represents the automatic segmentation results by the software embedded in our hospital (VIMS, Vitrea® 2), b) shows the automatic segmentation results by deep learning methods, c) shows the result of the edition based on b)). The blue dotted boxes highlight the significant differences among these three methods. We can observe that deep learning-based methods segment more accurately than the software. Furthermore, with manual refinement by experienced radiologists, we can correct some small breakage regions and obtain satisfactory airway segmentation results.
Airway Volume (AV)
The airway volume (AV) can be counted as below:
where xi denotes each voxel belongs to the airway.
Airway volume percent (AWV%)
The airway volume percent measures the percentage of airways occupying the lung volume. The formulation can be written as below:
where xi denotes each voxel belongs to the airway and the yi denotes each voxel belongs to the lung.
Total airway count (TAC)
The total airway count (TAC) denotes the number of branches by the segmented airway tree.
Airway fractal dimension（AFD）
The fractal complexity of the segmented airway luminal tree was measured by the Minkowski-Bougliand box-counting dimension . Precisely, the AFD can be calculated by the following formula:
where N stands for the number of sticks, ε stands for the scaling factor, and dimbox (Airway)stands for the fractal dimension.
Wall area percent (WA%)
Using 3D SLICER (http://www.slicer.org/), the wall area percent (WA%: (wall area/total bronchial area)×100) and the square root of wall area with an internal perimeter of 10mm (Pi10) were obtained as previously described[18, 19]. The mean WA% was calculated as the average of six segmental bronchi in each subject.
Pulmonary Function Tests
Pulmonary function tests (PFT) were performed on computerized spirometers (Masterscreen, Jaeger, Hochberg, Germany) according to the ATS/ERS recommendations[20, 21]. Each spirometer was calibrated daily, and a minimum of three satisfactory slow and forced vital capacity manoeuvers were required of each subject. All the data of PFT enrolled in this study were measured without using bronchodilators. The Basic information concerning age (years), height (cm), weight (kg), body mass index (BMI; kg/m2) were collected before PFT. Forced expiratory volume in 1 second (FEV1), forced vital capacity (FVC) and their ratios (FEV1/FVC), expiratory reserve volume (ERV), and vital capacity (VC) were acquired. A restrictive ventilation disorder was defined as FEV1/FVC >0.70 and FVC <80% predicted. FEV1/FVC <92% of predicted value was identified as an obstructive ventilation disorder, recommended by the guidelines for PFTs published by the Chinese Thoracic Society.
The SPSS 26.0 software (IBM Corporation, Chicago, IL, USA) and GraphPad Prism 8.40 (GraphPad Software Inc., San Diego, CA) were utilized to perform statistical tests and plot charts. Data were tested for normality using the Shapiro-Wilk test and non-parametric test if not normally distributed. One-way analysis of variance (ANOVA) or Kruskal-Wallis test was used to compare differences in demographics, pulmonary function tests, and CT measurements for age groups and pulmonary ventilation groups. Spearman (rs) correlation analysis was used to estimate associations between variables.
Multivariable models were generated using the enter approach to determine variables with significant associations with FEV1, FEV1/FVC, FVC. All variables were log-transformed. A P value < 0.05 was considered a statistically significant difference.