Upper Airway Volume Predicts Brain Structure and Cognition in Adolescents

Abstract

Rationale: One in 10 children experiences sleep-disordered breathing (SDB). Untreated SDB is associated with poor cognition, but the underlying mechanisms are less understood. Objectives: We assessed the relationship between magnetic resonance imaging-derived upper airway volume and children's cognition and regional cortical gray matter volumes. Methods: We used 5-year data from the Adolescent Brain Cognitive Development study (N = 11,875 children; 9-10 yr old at baseline). Upper airway volumes were derived using a deep learning model applied to 5,552,640 brain magnetic resonance imaging slices. The primary outcome was the Total Cognition Composite score from the NIH Toolbox (NIH-TB). Secondary outcomes included other NIH-TB measures and cortical gray matter volumes. Measurements and Main Results: The habitual snoring group had significantly smaller airway volumes than nonsnorers (mean difference, 1.2 cm³; 95% confidence interval [CI], 1.0-1.4 cm³; P < 0.001). Deep learning-derived airway volume predicted the Total Cognition Composite score (estimated mean difference, 3.68 points; 95% CI, 2.41-4.96 points; P < 0.001) per one-unit increase in the natural log of airway volume (∼2.7-fold raw volume increase). This airway volume increase was also associated with an average 0.02-cm³ increase in right temporal pole volume (95% CI, 0.01-0.02 cm³; P < 0.001). Similar airway volume predicted most NIH-TB domain scores and multiple frontal and temporal gray matter volumes. These brain volumes mediated the relationship between airway volume and cognition. Conclusions: We demonstrate a novel application of deep learning-based airway segmentation in a large pediatric cohort. Upper airway volume is a potential biomarker for cognitive outcomes in pediatric SDB, offers insights into neurobiological mechanisms, and informs future studies on risk stratification.

Keywords: airway volume; cognition; deep learning; pediatric; sleep-disordered breathing.

Figure 1.

Upper airway segmentation pipeline using deep learning. (A) Extent of the upper airway of interest in children with sleep-disordered breathing. The standard vertical air column extends from the superior turbinate to the inferior endplate of the C4 vertebra. Schematic representations of various severities of upper airway collapse are also shown. (B) Stack of T1-weighted magnetic resonance imaging slices and associated three-dimensional reconstruction. (C) Extent of the upper airway detected by the trained deep learning model. (D and E) Deep learning segmentation results for the upper airway on the same child in Year 1 (D) and Year 3 (E). Segmentation colors are manual expert (green), deep learning model (red), and overlap (yellow). Segmentation performance is shown across deep learning models trained using (F) 70 samples, (G) 40 samples, (H) 20 samples, (I) 10 samples, and (J) 5 samples. Segmentation colors are as above. (K) Dice coefficient by the scan number (n = 30) for the validation sample of 15 children (held-out testing set). (L) Average Dice score for various training dataset sizes, ranging from 5 to 70, showing comparable performance across sizes. (M) Analysis of voxel volume distribution and Dice metric performance illustrates the relationship between the number of training samples and the Dice metric, fitted with a power law model to predict performance for all 21,649 volumes. (N) Smoothed density estimate showing the spread and density of the voxel volume distribution for 70 training volumes. (O) Smoothed density estimate of the voxel volume distribution for all 21,649 training volumes.

Figure 2.

Deep learning segmentation model for measuring upper airway volume in children. (A) Mean and 95% confidence estimates for the difference between predicted and manually segmented airway volumes (expressed as cubic centimeters). (B) Boundaries of the upper airway of interest annotated by an expert (green) and detected by the deep learning model (red), and the highlighted overlap (yellow). (C) Relationship between airway volumes and age grouped by children’s snoring frequency as reported by the parents. Children’s airway volumes increased over time. Snoring was defined as habitual if they snored three nights or more per week, nonhabitual if less than that threshold, and none if they denied any snoring. (D) Difference in airway volumes grouped by snoring frequency. Habitual snorers had the smallest airway volumes across all waves of the study. Linear fits and 95% CIs are shown for C and D. CI = confidence interval.

Figure 3.

The relationship between airway volumes and longitudinal NIH Toolbox (NIH-TB) domain scores. Hexagonal bins are color coded by the sample size for each bin. Rows of panels represent data from Years 1, 3, and 5, respectively. The columns represent the data for the age-corrected NIH-TB domain scores. Linear fits and accompanying 95% CIs are also shown. CI = confidence interval; cryst = crystallized.

Figure 4.

Magnetic resonance imaging–derived upper airway volume predicts cognitive performance in children. (A–C) Cumulative distribution functions show the differences in airway volumes grouped by time (Years 1, 3, and 5), sex assigned at birth (male or female), and race (White, Black, or other). Results of statistical tests are shown in the violin/boxplot inset. Of note, Black children had, on average, airway volumes ∼10% smaller than White children. (D–M) Positive relationship between logn(airway volume) and children’s age-adjusted cognitive test scores measured by the NIH-TB, listing the sample size for each wave of data. Each line plot and the accompanying 95% confidence interval represents the marginal association of airway volume with age-adjusted cognitive test scores. Each model controls for the fixed effects of sex, race, type of testing (in-person, remote, or hybrid), body mass index z-score, socioeconomic status represented by total household income and educational attainment, and the random effects of site and subject identifier. (N) Coefficient for airway volume for each model. The point estimates and the accompanying confidence intervals indicate that airway volume predicted almost all cognitive test scores. cryst = crystallized; pic. vocab = picture vocabulary.

Figure 5.

Regional gray matter volumes mediate the relationships between upper airway volume and cognitive performance in children. (A) Cortical regions of interest (ROIs) associated with natural log-transformed airway volume measured using deep learning. To derive these plots, we assessed the relationship between upper airway volumes and cortical ROIs with a linear mixed effects regression model that included standard covariates as described in the text. The colored regions represent ROIs with different degrees of statistical significance shown in the scale bar. They are ranked by negative log-transformed P values. (B) Top five ROIs ranked by coefficients. (C) Lack of a relationship between airway volume and whole brain gray matter volume. However, ROIs in the frontal and temporal lobes, such as the orbital gyri and temporal poles, demonstrate strong relationships after adjustment for standard covariates in linear mixed effects regression models. (D) Top 10 ROIs that mediate the relationship between log-transformed airway volume and NIH-TB domain scores, ranked by the proportion mediated (percentage). (E) Each mediation model as shown in controls for the fixed effects of age, sex, race, total household income, highest parental education, body mass index z-score, and the random effects of site and scanner. ant = anterior; cing = cingulate; cryst = crystallized; G = gyrus; L = left; pic. vocab = picture vocabulary; R = right; S = sulcus; STG = superior temporal gyrus; temp = temporal.