Patient‐Specific Computational Fluid Dynamics Modeling of Airway Collapse in Infants With Robin Sequence Authors: Gautam Jalori, Michael Barbour, Michael Bindschadler, Kelly Evans, Jake Dahl, Alberto Aliseda Published: 2026 ABSTRACT ABSTRACTRobin Sequence (RS) is a congenital condition in which patients experience dynamic, periodic obstruction or collapse of the upper airway due to an underdeveloped jaw and a posteriorly displaced tongue. Current clinical techniques for evaluating airway obstruction do not provide quantifiable data and fail to account for the dynamic nature of obstruction or collapse. There is no standardized criterion to characterize obstruction or collapse severity. This study presents the first method that extracts airway motion from 4‐dimensional computed tomography and performs a patient‐specific, moving‐mesh computational fluid dynamics (CFD) analysis of RS patients with complete airway collapse or obstruction. To quantify the effects of airway collapse, airflow dynamics are analyzed using both instantaneous metrics (velocity, pressure, and energy dissipation rate) and cycle‐averaged metrics (resistive work of breathing). These results are compared between a collapsing airway and its synthetic non‐collapsing counterpart. To validate the synthetic non‐collapsing case, it is further compared with a patient‐specific non‐collapsing airway. The results show that, to achieve the same tidal volume, the collapsing case requires significantly greater computed breathing effort (7.55 mJ/cycle) than the synthetic non‐collapsing case (3.68 mJ/cycle). The shorter inspiration time due to the collapse leads to a higher inlet velocity, resulting in 1.7 times the maximum velocity during peak inspiration (just prior to collapse) and a 2.7‐fold higher pressure drop in the collapsing case compared to the synthetic non‐collapsing case. At the onset of collapse, a sharp spike in localized energy dissipation rate is observed due to the abrupt deceleration and dissipation of peak flow velocities. This methodology provides a novel approach to understanding the airflow dynamics of RS patients with airway collapse. It enables quantitative comparison between collapsing and non‐collapsing airways, thereby offering the potential to support more informed and objective clinical decision‐making. Introduction Robin Sequence (RS) is a congenital condition characterized by an underdeveloped jaw (micrognathia), backward displacement of the tongue (glossoptosis), and resultant upper airway obstruction (UAO) [1]. The birth prevalence ranges between 1 in 8000 and 1 in 14,000 births [2]. The mortality rate for children with RS is reported to be between 2.2% and 26% [3, 4, 5]. Treatment for this condition varies from conservative approaches, such as prone positioning and nasopharyngeal airway placement, to complex invasive surgeries such as mandibular distraction osteogenesis (MDO), tongue‐lip adhesion, and tracheotomy [1, 6]. The Oral Airway Plate (OAP), a device consisting of an intraoral acrylic appliance with velar extension, has recently been introduced as a safe and successful non‐surgical treatment modality though it is not yet widely offered in the US [7, 8, 9, 10, 11]. There is a lack of standardized criteria for classifying the severity of airway obstruction, leading to variability in treatment decisions. Current evaluation techniques, such as endoscopy and nasopharyngoscopy, help visualize tongue position in the oral cavity and oropharynx but do not provide quantifiable data. Standard CTs are static, and as there is significant upper airway motion during respiration in RS, a single image is unlikely to effectively capture obstruction severity [12]. Neither modality provides information on functional respiration or airflow. Because of the diversity in the phenotype of RS, each treatment should be tailored to the individual patient based on the etiology and mechanism of airway obstruction [1]. Interpatient differences in morphology and functional respiration must be understood and carefully assessed to effectively risk‐rank patients and identify the optimal treatment modality [13]. In recent years, the computational fluid dynamics (CFD) method has been widely employed to analyze the airflow in both healthy and diseased human airways [14]. Most of the studies of CFD of the upper airway (UA) are related to Obstructive Sleep Apnea (OSA). The analysis in these original studies relied on data from static CT images and employed stationary airway boundaries, neglecting the relationship between airway wall motion and airflow dynamics [15, 16]. Some studies have used FSI simulations to incorporate UA motion in CFD simulations [17, 18, 19, 20], but these simulations do not account for wall motion driven by neuromuscular activity (wall motion is only driven by flow‐based pressure changes) and often assume uniform structural properties, which are heterogeneous both within a single patient and across populations. More recently, to bridge these gaps, dynamic image modalities, such as cine MRI or 4DCT images have been combined with CFD simulations, incorporating airway wall motions [21]. In a study comparing flow through a dynamic airway with motion defined by cine‐MRI, and its static counterpart, variations of 10% and 20% were found between inspiratory and expiratory resistance % [22]. Gunatilaka et al. [23] showed that the resistive component of work of breathing was increased by an average of 337% ± 295% in the dynamic simulation as compared to the static simulation. The static simulation was performed on the largest open airway, resulting in the highest possible difference in results between the dynamic and static simulation. To date, no studies have incorporated patient‐specific airway wall motion throughout an entire respiratory cycle to evaluate airflow dynamics in infants with RS. Existing CFD studies have analyzed flow dynamics using static upper airway geometries in RS patients. Marzano‐Rodrigues et al. [24] used 3D segmentations derived from static cone‐beam CT to characterize airflow severity in six young adults with RS, and Barbour et al. [25] incorporated 4DCT derived 3D segmentation at peak inspiration and expiration to analyze flow metrics of infants with RS (the present work builds upon that initial study). However, those approaches did not capture the unsteady airflow induced by airway wall motion or the transient dynamics associated with UA collapse. Moreover, while both OSA and RS patients experience events of complete collapse/apnea in the airways [13, 26], the dynamics related to complete flow cutoff in the airway have not been studied. Hereafter, the term “collapse” refers specifically to complete airway collapse. In this study, we develop a patient‐specific CFD model integrated with 4DCT‐derived airway motion to quantify airflow dynamics throughout the full respiratory cycle in an infant with RS. Airway collapse was implemented using resistance‐based porous region and interfaces at the collapse site that impose time‐varying flow restrictions, enabling a gradual transition from unobstructed flow to complete flow shut‐off without explicit geometric closure. By explicitly modeling airway collapse, we extract instantaneous and cycle‐averaged metrics, including regional work of breathing, to assess collapse severity. Results are compared against a synthetic, non‐collapsing case derived from the same anatomy and airway motion, without the complete collapse phase (in which the synthetic case maintains a narrowed but open airway at the smallest cross sectional area that is resolved fully from the 4DCT and from which the collapse model is applied), highlighting the qualitative and quantitative changes introduced by apnea events, which demonstrate the critical importance of modeling airway collapse in the image‐based computational assessment of RS severity (and other related pathologies that include UA collapse). To validate the results of the synthetic non‐collapsing case, we also compare results with a patient‐specific non‐collapsing case, confirming that the synthetic model provides a realistic baseline to evaluate flow dynamics of a collapsing airway. Method Patient Selection and Imaging 4D‐CT data was collected from a 54‐day‐old patient, weighing 3.07 kg, with approval from Seattle Children's Hospital IRB. Images were obtained with a dual‐source CT scanner (Somatom Force, Siemens, Erlagen Germany). The gantry rotation speed was set to 0.25 s and 3D images were reconstructed every 0.1 s. Images were acquired for 4 s. The image extent covered the frontal sinus down to the subglottis. Spatial resolution was 0.33 mm and slice thickness was 0.6 mm. The patient was placed in a supine position in the scanner bed. Supplemental oxygen (< 5 lpm) was provided through a low‐flow nasal cannula based on the patient's need. SPO2 levels were monitored with pulse‐oximetry. Supplemental O2, if being provided, was briefly turned off for the 4 s of scan acquisition. To synchronize CT acquisition with respiratory phase data, a DSLR camera was pointed at the CT scanner bed, focused on the patient's chest, while maintaining the CT indicator light in the field of view. Vertical chest motion was computed using custom software and time‐locked to the CT scan using the CT scanner indicator light. The gradient of the chest motion curve provides an independent measure of the timing of inspiration and expiration with respect to the CT‐image acquisition, allowing for the synchronization of flow rate and dynamic airway anatomy in time, as shown in Figure 1 (1 s cutout from around 4 s of data). A sinusoidal curve is fitted to the motion gradient data to estimate the inspiration and expiration phases. The highest motion gradient is set to correspond to peak inspiration and the lowest motion gradient to peak expiration. Gradients approaching zero are set to time the transition between the two phases. From the fitted curve, the respiratory cycle length was found to be 1 s with equal inspiratory and expiratory durations. The amplitude of this sinusoidal flow waveform is calculated from an empirical tidal volume relation discussed in Section 2.3. The patient underwent complete UA collapse periodically in every respiratory cycle, with collapse occurring at 0.3–0.5 s during inspiration. Chest motion and its gradient, with mass flow mapped onto the gradient curve. All the images in the 4D‐CT are segmented (three shown in Figure 2) using in‐house tools developed within the open‐source software 3D Slicer (slicer.org [27]) for the purpose of initial anatomic analysis, including spatial and temporal changes in the airway cross‐sectional area and airway volume. A volume (shown in Figure 2) is added to the 3D segmentation, connecting the mouth and nostrils into a single fluid zone that acts as the ambient boundary condition in the simulation. Using the data shown in Figure 1, a single 3D segmentation is extracted corresponding to the start of the flow rate curve. This segmentation serves as the reference or initial fluid domain to which the airway wall motion is applied. To extract airway motion from the 4DCT images, deformable registration is performed on 11 images over the duration of 1 s (respiratory cycle length) in the 3D Slicer software. Using the Sequence Registration module and underlying Elastix registration toolbox [28, 29] fixed image volume (corresponding to the initial segmentation or fluid domain) is registered to the sequential image volumes. Registration was performed with multi‐resolution B‐spline transformation, setting the grid spacing to 4 voxels. Optimization was carried out using adaptive stochastic gradient descent with random spatial sampling to improve convergence efficiency. Cubic B‐spline interpolation was applied during both optimization and final resampling to ensure smooth and physically plausible deformation fields. As a result, a transform sequence (3D displacement field varying in time) and a motion‐compensated volume sequence are obtained, with the transform sequence used to deform the segmented volume (reference geometry) (Figure 3) with the help of control points. 3D segmentation and sagittal CT image (cutting through the collapse) at time instance (a) before collapse (0.2 s), (b) of collapse (0.4 s), (c) after collapse (0.6 s). (a) Flow domain with boundary conditions and initial set of control points, (b) anatomical regions. Control points (shown in Figure 3) were extracted from the reference geometry using a Poisson‐disk distribution in MESHLAB. After experimenting with this tool, we found a sample size of 1000 gave the best results, as a compromise between excellent spatial resolution and careful allocation of computing resources and time. In Slicer, this reference set of control points was duplicated and deformed using the transform sequence obtained from Elastix‐based sequence registration. For each time frame, the corresponding 3D displacement transformation was computed and applied to a cloned instance of the control points set, yielding a deformed control point configuration for that time point. Using a Python script, this process was repeated across the 10 frames of the respiratory cycle, while preserving the reference, undeformed, control points set. The resulting time‐indexed control points sets were exported in FCSV (Fiducial Comma Separated Value) file format (used to store 3D coordinates in Slicer) for subsequent use in the deforming‐mesh simulation. To prescribe motion at every time step of the solver, the temporal resolution of the control points sets was refined from the imaging frame rate of 0.1 s to the time intervals in the CFD solver, dt = 0.001 s, by using a periodic B‐spline interpolation. To reproduce the periodic nature of the respiratory cycle, the final set of control points was replaced with the initial set and periodic boundary conditions were imposed in the temporal interpolation. In addition to the interpolated control point coordinates, incremental displacements between successive control point sets were computed and provided to the mesh morpher, and incremental displacements were used to apply the deformation. Solver and Meshing Details The Navier–Stokes equations were solved with a finite‐volume segregated flow solver (Star‐CCM+), using a second‐order implicit method for time advancement with a time step of 0.001 s. Spatial discretization is performed using a third‐order MUSCL scheme with a minmod gradient limiter (bounding difference = 0.15), and the realizability option is disabled. We employ the Unsteady Reynolds Averaged Navier–Stokes (URANS) framework, with a k‐ω SST turbulence model, including Low Reynolds Damping. This turbulence model has shown in our work (see Supporting Information) to provide the minimal discrepancy between computed and budgeted energy dissipation in the volume, as well as optimum agreement with experimental data in the literature [15, 30]. All numerical settings were selected based on an energy dissipation study. In this study, the terms of the energy equation [31] are evaluated and integrated over one respiratory cycle. After neglecting the shear work and fluctuating kinetic energy term (which vanish when integrated over a full respiratory cycle), the governing balance reduces to: (1) ∫Si∪So∪SwP+ρv22(v.n)dS+μ2∫VD:DdV+∫VρkωdV=0 where Si,So,Sw denotes the inlet, outlet and the wall surfaces, P is the static pressure, ρ is the fluid density, v is the velocity vector, n is the normal vector, μ is the dynamic viscosity, V is the volume, and the strain rate tensor defined as D = ∇v+(∇v)T. The added term “kω” represents turbulent dissipation, where k is the turbulent kinetic energy and ω is the specific dissipation rate. The solver settings were adjusted to ensure this energy balance is satisfied. The first term, integrated over the respiratory cycle, is used to compute the Work of Breathing (WOB) between selected surface pairs [25, 32]. WOB estimated is the resistive component of overall WOB, as the value calculated does not include work done to overcome the elastic recoil of the lung and the chest. A polyhedral mesh is used with approximately 3 million cells, fifteen prism layers were prescribed along the airway walls with a total thickness of 0.4 mm, and ten prism layers in the outlet and nasal cavity regions with a total thickness of 1.2 mm, maintaining wall y+ < 1 throughout the respiratory cycle. A mesh independence study was conducted using base mesh sizes of 0.5, 0.4, 0.3, 0.25, and 0.2 mm, with maximum velocity, pressure, and energy dissipation monitored as convergence metrics. The variability in maximum velocity decreased from approximately 6% at 0.5 mm to 1.5% at 0.25 mm. In addition, the variability in pressure during peak inspiration decreased from 2.9% at 0.5 mm to 0.01% at 0.25 mm. Although velocity‐based metrics showed limited improvement beyond 0.25 mm, the residual in the energy balance (Equation 1) decreased from 14% to 11% when the mesh was refined from 0.25 mm to 0.2 mm. Based on this improvement in satisfying the energy balance, the 0.2 mm mesh was selected for all simulations. A time step of 1 ms was selected based on a time‐step independence study, the difference in peak inspiratory maximum velocity was 0.8% compared with a simulation performed using a finer time step of 0.1 ms. Collapse Implementation and Boundary Conditions The CT images for three time points in the respiratory cycle, t = 0.3, 0.4 (Figure 2b), and 0.5 s, present minimal cross‐sectional area at the narrowest point. These three time points represent the phase of inspiration in which the airway is considered to be collapsed. The collapse is identified from 4DCT images at the time when the distance between the opposite points of the airway cross‐section, at its minimum opening, is comparable to the image resolution. Given that the airway is closed, we model this segment as a resistive element since it would be unphysical to try to impose a certain tidal flow through a geometrically collapsed cross‐section. This results in a fast, but progressive shut‐down in flow, in a manner that replicates quantitatively the decreasing flow in an increasingly resistive infinitesimal cross section airway. We developed a new methodology to include the effect of collapse in the airway flow simulation that is robust to image segmentation uncertainty. First, we replace the control points sets computed at the times when collapse is observed (t = 0.3, 0.4, 0.5 s in this case), by a new set generated by interpolating between the control points just before, t = 0.2 s, and just after, t = 0.6 s. To imitate the collapse in the airway, we use the porous region and interface functionality of STARCCM+. A porous region is defined in the flow domain at the site of collapse in the 4DCT images and the planes between the porous and the normal flow domain are the upstream and downstream interfaces (shown in Figure 3). At the porous region, the value of flow resistance can be defined in the full range from zero resistance to infinity (complete flow cut‐off). The interface is set to internal by default, acting as a zero‐resistance plane. Resistance in the porous region is gradually ramped up during collapse to achieve nearly zero flow; however, to achieve complete flow cut‐off, rather than prescribing very high resistance values (to avoid solver instabilities), a baffle model at the interface was used, which acts as a wall and shuts off the flow. This approach does not explicitly resolve the small, short‐lived, flow structures caused by the realistic movement of the collapsing wall, at millisecond and sub‐millimeter scales, but captures its overall effect on the airway flow. The porous region was activated only to increase and decrease the flow resistance before and after the collapse (or baffle interface), respectively; for the rest of the duration, the porous region was switched to a normal flow domain. The temporal resolution of the 4DCT images is 0.1 s, which prevents the identification of the exact onset and recovery times of airway collapse. Therefore, the porous region was activated 0.05 s after the last image showing an open airway, corresponding to the midpoint between consecutive image frames. A sensitivity analysis using activation times of 0.03 s and 0.07 s after the last open‐airway frame resulted in only minor changes in the cycle‐integrated work of breathing, indicating that the results are not sensitive to estimation of the exact collapse timing. Here is the timeline of models used for this interface: Internal Interfaces On, Porous Region Off (0–0.25 s): to have a normal flow without any added obstruction/resistance during inspiration.Internal Interfaces On, Porous Region On (0.25–0.3 s): to imitate the closing of the airwayUpstream Baffle Interface On, Downstream Internal Interface On, Porous Region Off (0.3–0.5 s): to imitate a complete flow shut‐off or airway collapseInternal Interfaces On, Porous Region On (0.5–0.525 s): to imitate the opening of the airwayInternal Interfaces On, Porous Region Off (0.525–1.0 s): to again have a normal flow without any added obstruction/resistance during expiration. Internal Interfaces On, Porous Region Off (0–0.25 s): to have a normal flow without any added obstruction/resistance during inspiration. Internal Interfaces On, Porous Region Off (0–0.25 s): to have a normal flow without any added obstruction/resistance during inspiration. Internal Interfaces On, Porous Region On (0.25–0.3 s): to imitate the closing of the airway Internal Interfaces On, Porous Region On (0.25–0.3 s): to imitate the closing of the airway Upstream Baffle Interface On, Downstream Internal Interface On, Porous Region Off (0.3–0.5 s): to imitate a complete flow shut‐off or airway collapse Upstream Baffle Interface On, Downstream Internal Interface On, Porous Region Off (0.3–0.5 s): to imitate a complete flow shut‐off or airway collapse Internal Interfaces On, Porous Region On (0.5–0.525 s): to imitate the opening of the airway Internal Interfaces On, Porous Region On (0.5–0.525 s): to imitate the opening of the airway Internal Interfaces On, Porous Region Off (0.525–1.0 s): to again have a normal flow without any added obstruction/resistance during expiration. Internal Interfaces On, Porous Region Off (0.525–1.0 s): to again have a normal flow without any added obstruction/resistance during expiration. Currently, there is no validated and safe method to measure flow rate during respiration in infants undergoing a 4D‐CT scan, especially those with UAO. Therefore, we impose boundary conditions based on an estimate of minute‐volume (volume of air inhaled or exhaled in 1 min) based on the patient's weight via the following empirical relation, and a sinusoidal temporal waveform for the mass flow curve to match the time‐integrated minute‐volume total flow and the duration of the respiratory cycle measured from the 4D‐CT image sequence: (2) mv(milliliters)=250*w(kg); where w is the weight of the patient (3.07 kg), and mv is the minute‐volume [33]. Using the above relation and the respiration period (1 s) from the 4DCT data, the tidal volume was found to be 12.8 mL. The flow waveform is synchronized according to the chest movement data, and because the inspiratory and expiratory phase times are equal in this particular patient, stretching of the harmonic waveform was not required (Figure 1). This waveform is used to calculate the maximum velocity as a function of time at the inlet, and from this maximum velocity at the centerline of the airway, a Womersley flow velocity profile is prescribed as an inlet boundary condition at the trachea of the geometry for non‐collapsing cases. The tracheal inlet in the segmented anatomy is extended (as shown in Figure 3) in VMTK [34] to have a circular inlet cross‐section where the velocity profile boundary condition can be applied. To utilize the porous region and interface functionality for collapsing cases, inflow velocity boundary conditions cannot be applied, as they would violate conservation of mass (and the physics of the problem). To maintain physically and physiologically realistic conditions in the simulations with collapse, pressure boundaries are used at the inlet and outlet of collapsing cases. To get pressure values and maintain the desired tidal volume, three additional intermediate simulations are performed before the final one with a no‐slip boundary condition at the airway walls for all four simulations. In the first intermediate simulation, the complete collapse event is intentionally ignored, and a simulation is created as if the contracting airway motion pauses between 0.2 and 0.6 s (the timepoints on either side of collapse), but complete collapse doesn't occur. As described earlier in this section, the airway motion field between 0.3 and 0.5 s, when the airway is deemed completely collapsed, is computed by interpolating between the geometries at 0.2 and 0.6 s. In this synthetic case, which is also presented in the results for comparison, the sinusoidal mass flow curve is prescribed at the inlet boundary as shown in Figure 4 (synthetic), where the outlet is set to ambient pressure, the porous region is never activated, and the interface boundary is set to internal interface. The resultant inlet pressure is then calculated. To model the collapsing case, the inlet boundary condition is changed to a pressure inlet, using pressure calculated in the synthetic case when inlet velocity is applied at the boundary. However, in this second stage simulation for the collapsed case, the resistance in the porous region increases from zero at 0.25 s to infinity (baffle interface) at 0.3 s. This causes the cumulative volume flow during inspiration to be lower than in expiration. To correct for this, a third stage simulation is run with flow inlet boundary conditions, but this time with the volume flow in inspiration scaled up such that the inspiratory and expiratory volumes are equal, accounting for the time during inspiration when the airway is collapsed. In the third simulation, the scaled mass flow curve is prescribed at the inlet boundary and pressure values are recorded and then applied as the inlet boundary conditions in the final (fourth) stage of the simulation with collapse. Figure 4 shows the mass flow for the collapse case (final simulation) with pressure boundary conditions that replicated minute‐flow conditions. There is currently no validated and safe method to directly measure the pressure within the trachea during collapse; therefore, we assume that the pressure drop during the final simulation is equivalent to that obtained in the third, intermediate simulation during the corresponding period of collapse. Mass Flow at inlet of collapsing, synthetic and non‐collapsing case. To establish the physiological relevancy of the synthetic non‐collapsing results, we compare it with a patient‐specific non‐collapsing case diagnosed with RS. For this purpose, we consider a 23‐day‐old patient weighing 4.4 kg, with a respiratory duration of 1.1 s and a resultant tidal volume of 20.17 mL. The image acquisition details, calculation of the wall displacement field, mesh‐morphing and simulation parameters are the same as described in 2.1 and 2.2, except that the pre‐simulation setup for the non‐collapsing case does not include a porous region, as the tidal flow is driven through the constricted cross‐section. Also, no images are excluded from the motion field calculation as there is no collapse event. A single simulation is performed using a velocity inlet boundary condition, as described above, where the inlet velocity is calculated from the mass flow rate obtained from the chest movement data and the empirical tidal volume relation. Hereafter, the synthetic non‐collapsing case is referred to as the “synthetic case,” and the patient‐specific non‐collapsing case is referred to as the “non‐collapsing case.” Viscous Resistance (Rv) was applied in the porous region to impose an additional pressure drop across the region according to ΔPregion=RvVL where V is the velocity in the porous region and L is the length of the region along the flow. Rv is determined such that the imposed pressure drop ΔPregion is equivalent to the radius of the porous region collapsing to zero in a cosine‐based temporal profile. Let r(t) denote the radius of the porous region equivalent to the collapsing airway during the transition period 0.25≤t≤0.30s. Rv is calculated as follows: The velocity in the porous region in the transition period 0.25≤t≤0.30s is computed via a Poiseuille‐based (quadratic) relationship. (3) V(t)=CregionΔPoff(t)r(t)2 where V(t) is the velocity in the porous region, and ΔPoff(t) is the pressure drop across the porous region obtained from the simulation with the porous resistance deactivated. Cregion is calculated using known velocity and pressure drop across the porous region at 0.25 s.The pressure drop across the airway required to drive the computed velocity in the porous region is estimated as (4) ΔPeffective(t)=CairwayV(t)rsim(t)2 where rsim(t) is the actual radius of the porous region in the simulation. Cairway is calculated using known velocity and pressure across the airway at 0.25 s.The pressure drop that must be explicitly applied across the porous region is then (5) ΔPregion(t)=ΔPBC(t)−ΔPeffective(t) where ΔPBC is the boundary condition pressure drop across the airway.Viscous Resistance (Rv): (6) Rv(t)=ΔPregion(t)LV(t) The velocity in the porous region in the transition period 0.25≤t≤0.30s is computed via a Poiseuille‐based (quadratic) relationship. (3) V(t)=CregionΔPoff(t)r(t)2 where V(t) is the velocity in the porous region, and ΔPoff(t) is the pressure drop across the porous region obtained from the simulation with the porous resistance deactivated. Cregion is calculated using known velocity and pressure drop across the porous region at 0.25 s. The velocity in the porous region in the transition period 0.25≤t≤0.30s is computed via a Poiseuille‐based (quadratic) relationship. (3) V(t)=CregionΔPoff(t)r(t)2 where V(t) is the velocity in the porous region, and ΔPoff(t) is the pressure drop across the porous region obtained from the simulation with the porous resistance deactivated. Cregion is calculated using known velocity and pressure drop across the porous region at 0.25 s. The pressure drop across the airway required to drive the computed velocity in the porous region is estimated as (4) ΔPeffective(t)=CairwayV(t)rsim(t)2 where rsim(t) is the actual radius of the porous region in the simulation. Cairway is calculated using known velocity and pressure across the airway at 0.25 s. The pressure drop across the airway required to drive the computed velocity in the porous region is estimated as (4) ΔPeffective(t)=CairwayV(t)rsim(t)2 where rsim(t) is the actual radius of the porous region in the simulation. Cairway is calculated using known velocity and pressure across the airway at 0.25 s. The pressure drop that must be explicitly applied across the porous region is then (5) ΔPregion(t)=ΔPBC(t)−ΔPeffective(t) where ΔPBC is the boundary condition pressure drop across the airway. The pressure drop that must be explicitly applied across the porous region is then (5) ΔPregion(t)=ΔPBC(t)−ΔPeffective(t) where ΔPBC is the boundary condition pressure drop across the airway. Viscous Resistance (Rv): Viscous Resistance (Rv): A similar process is used to calculate the viscous resistance coefficient applied after the collapse duration is over to gradually open the flow. Results Patients‐Specific Versus Synthetic Non‐Collapse Using 4DCT and CFD, a patient‐specific case with a collapsing airway, a synthetic, and a non‐collapsing case were successfully modeled, and results were extracted and analyzed. Figure 5 compares the result of the synthetic and non‐collapsing case. The top plot shows the progression of maximum velocity and cross‐sectional area (CSA) with normalized time in a full respiration cycle, with inspiration in the first half and expiration in the second. At the constriction, the synthetic case exhibits a maximum velocity of 22 m/s, which is nearly twice that of the non‐collapsing case. This increase is primarily driven by 2.5 times lower CSA in the synthetic case. However, 1.6 times lower tidal volume reduces the velocity increase to approximately 2‐fold rather than 2.5‐fold. However, the WOB/cycle was nearly identical in both cases, with values of 3.5 mJ for the non‐collapse case and 3.68 mJ for the synthetic case. This can be explained by the higher tidal volume in the non‐collapsing case, leading to higher inlet velocities, as shown in the bottom plot. The maximum inlet velocity at peak inspiration is almost 1.65 times higher in the non‐collapsing case, while the corresponding pressure is nearly half that of the synthetic case. Although the synthetic case has a smaller CSA, the non‐collapsing case requires comparable breathing efforts due to the higher inhaled volume. Having established that the synthetic case exhibits airflow dynamics within the range of a real non‐collapsing patient, the following results present a comparison of the collapsing case and the non‐collapsing, synthetic equivalent. Top plot: Maximum velocity and CSA at site of constriction of the synthetic and non‐collapsing case, bottom plot: Maximum velocity and average pressure at the inlet of the synthetic and non‐collapsing case. Patients‐Specific Collapse Versus Synthetic Non‐Collapse Because physiological parameters such as airway anatomy, respiratory period, and tidal volume vary across subjects, the synthetic case provides a common baseline for evaluating collapse severity. Velocity The evolution with time of the maximum velocity and cross‐sectional area in the porous region and the trachea is shown in Figure 6. Prior to collapse, the collapsing case exhibits substantially higher velocities in the porous region compared to the synthetic case, with peak velocities reaching approximately 39 m/s at 0.25 s, compared to about 23 m/s in the synthetic case. The elevated velocity in the collapsing case arises because the same tidal volume is inspired over a shorter effective inspiration duration (almost half of the synthetic case), resulting in approximately 1.7 times increase in flow velocity. Although the geometric CSA of the porous region remains nearly constant throughout the simulation, the resistance‐equivalent CSA abruptly decreases to zero at approximately 0.3 s, reflecting the collapse imposed through a porous resistance and interface model. During expiration, both cases exhibit lower, more gradually changing, and identical velocity at the collapse (as indicated by the overlapping dashed and dash‐dotted lines) and at the trachea. However, higher velocities are observed during expiration at the site of collapse than at the trachea, indicating that constriction persists during this phase, if not equally severe as a collapse. Maximum velocity at trachea and site of collapse/constriction vs. time, for collapse and synthetic cases. Pressure The pressure response further highlights the severity of the resistance associated with airway collapse. As shown in Figure 7, the collapse case exhibits peak inspiratory pressures of up to −713 Pa, compared to −267 Pa in the synthetic case, corresponding to a 2.7‐fold increase in pressure drop across the constricted segment. This amplification is consistent with the elevated inspiratory velocities observed before collapse. The lowest pressure occurs shortly before collapse (around 0.25 s). In the synthetic case, pressure recovers as the cross‐sectional area expands at the narrowest point in the airway. For the collapsing case, the porous region pressure drop (dotted line in Figure 7), models the effect of the collapse airway, matching the inlet pressure during collapse (between 0.3 and 0.5 s) to yield zero flow through resistance‐driven flow cut‐off rather than the actual anatomical airway closure, which would be numerically unstable. At the end of inspiration (t = 0.5 s), the collapsed airway reopens, the porous region and interface model fades away (with zero added pressure drop), and the anatomical conditions with an open airway determine flow driven by the prescribed pressure at the boundary. Inlet pressure for the collapse and synthetic cases, collapse porous region pressure drop and airway volume vs. time. Peak Inspiratory Velocity 3D Renderings The higher pressure drop across the airway in the collapse case leads to the formation of a more intense inspiratory jet, as shown by the velocity profiles at peak inspiration in Figure 8. In the collapse case, this higher velocity jet persists downstream into the lower oropharynx and laryngopharynx. This sustained jet structure results in strong velocity gradients and more coherent turbulent structures, whereas the synthetic case exhibits weaker coherent structures that spread rapidly downstream of the constriction, leading to a broader velocity distribution and reduced peak velocities impinging on the laryngopharyngeal region. The cross‐sectional velocity contour plots, shown in Figure 8, further demonstrate the steeper velocity gradients and a more concentrated high‐velocity jet core in the collapse case, with stronger shear layers. The persistence of high‐velocity flow in the collapse case explains the elevated downstream viscous losses quantified in the regional work‐of‐breathing analysis in the following results. 3D renderings and velocity contours in cross‐sectional planes cutouts from flow domain at peak inspiration for (a) synthetic case and (b) collapse case. Energy Dissipation Rate and Work of Breathing To quantify the effort made in breathing, energy dissipation is calculated and integrated over the respiratory cycle, a variable called Work of Breathing (WOB) in the clinical literature. Due to the higher pressures and velocities reached during the shorter inspiration time needed to make up for the time when the airway is collapsed, the peak inspiratory energy dissipation rate in the collapsing case is nearly five times greater than in the synthetic case, as shown in Figure 9a. When integrated over the respiratory cycle, the total WOB increases from 3.68 mJ in the synthetic case to 7.55 mJ in the collapsing case, a 2‐fold increase in the energy required for breathing due to airway collapse. (a) Energy dissipation for the collapsing and synthetic cases in the entire flow domain throughout the respiratory cycle, (b) Energy dissipation for the synthetic case in the different anatomical regions throughout the cycle, (c) Energy dissipation for the collapsing case in the different anatomical regions throughout the cycle, (d) Region‐specific WOB during inspiration in the collapsing and synthetic cases, (e) Region‐specific WOB during expiration in the collapsing and synthetic cases. To identify the anatomical regions most affected by airway collapse, the airway anatomy was divided into distinct regions, as shown in Figure 3b. Figure 9b illustrates the temporal evolution of energy dissipation rate across these regions for the synthetic case. During inspiration, the constriction of the upper oropharynx forms a jet that propagates downstream into the lower oropharynx, inducing strong shear layers and flow fluctuations along with jet impingement on the oropharyngeal walls, which leads to the largest viscous losses and makes the lower oropharynx the dominant contributor to energy dissipation rate during inspiration. In contrast, the trachea exhibits a minimal energy dissipation rate due to its relatively larger cross‐sectional area and more uniform velocity profile. In the collapsing case, illustrated in Figure 9c, the highest energy dissipation rate is similarly observed in the lower oropharynx. As collapse initiates, flow is abruptly brought to a stop, causing energy dissipation rate to rapidly decrease to zero in regions downstream of the collapse. However, in the upper oropharynx, energy dissipation rate initially increases as the high‐velocity jet at the constriction undergoes sudden deceleration, producing intense local viscous losses before ultimately vanishing once the flow is cut off and the jet inertia dissipates. Figure 9d,e summarize the regional WOB for both collapsing and synthetic cases during inspiration and expiration, respectively. From these figures, it is clear that the lower oropharynx contributes most to the total WOB during inspiration, and the effect of the jet formed at the constriction is measured further downstream, in the laryngopharynx, where the WOB is slightly higher than in the upper oropharynx. On the opposite end, during expiration more energy is dissipated in the nasal cavity and upper oropharynx as the jet formed at the constriction in the oropharynx flows in the opposite direction, developing shear layers throughout, and eventually impinging on the wall of upper anatomical features close to the airway entrance. Discussion This study describes in detail a patient‐specific method based on 4DCT‐derived dynamic airway anatomy and wall motion, and CFD to model airflow dynamics across a full respiratory cycle in RS patients with UA obstruction or collapse. By combining a moving airway domain with a resistance‐based representation of collapse, the model enables quantification of both instantaneous (velocity, pressure, and dissipation spikes) and cycle‐averaged (WOB) flow metrics. To breathe the same tidal volume as the anatomy‐equivalent synthetic non‐collapsing case, twice the breathing effort is needed in the collapsing airway patient compared to the synthetic one. This is due to higher peak velocities leading to more intense jet formation, which in turn induces greater shear in the flow and greater viscous and turbulent energy dissipation. Analyzing purely geometric parameters, such as the minimum CSA and the airway volume, cannot accurately predict functional flow metrics, especially in a collapse case because the minimum CSA is zero. While collapse duration may offer additional anatomical information, it only quantifies the duration of flow interruption and does not capture the rate and dynamics of collapse, which govern transient flow acceleration, energy dissipation, and the resulting physiological impact of collapse. This limitation is reflected in clinical observations, Humphries et al. [35] studied upper airway morphological changes in infants with RS before and after MDO for 15 patients. An increase in volume of the retro‐palatal and retro‐glossal regions helped reduce Apnea‐Hypopnea Index (AHI); however, 6 patients still had abnormal AHIs even with significantly larger CSAs. It is unclear what the threshold for retropalatal or retroglossal space enlargement must be to overcome upper airway collapse and, thus, whether the amount of distraction required to achieve this result is feasible or practical, particularly in syndromic patients. By resolving airway pressures, flow rates, and collapse timing over the full respiratory cycle, CFD enables direct assessment of some of the mechanical drivers of collapse. In the present case, airway collapse occurred in phase with peak inspiratory pressure, indicating that elevated pressure loads (and not anatomy alone) may play a dominant role in initiating collapse. This interpretation is consistent with prior work by Xiao et al. [36], who quantified the power exchange arising from interactions between airflow‐induced pressure forces, neuromuscular activity, and airway wall motion. In patients with severe Obstructive Sleep Apnea (OSA), neuromuscular forces dominate the first half of inspiration, causing dilation, while pressure forces prevail in the second half, causing constriction. These results also demonstrate why reliance on peak quantities (Table 1) alone can be misleading when comparing collapse severity. In this patient, peak velocity at the porous region suggests that the collapsing airway is approximately 1.7 times as severe as the synthetic case, peak inspiratory pressure indicates a 2.7‐fold increase in severity, and peak energy dissipation rate implies an amplification of nearly 5‐fold. Each of these metrics captures only instantaneous information and fails to account for the duration over which elevated metrics are sustained. By resolving the full respiratory cycle and integrating energy dissipation rate over time, work of breathing (WOB) provides a cycle‐averaged measure of respiratory effort. Using this metric, the collapsing case is found to require twice the effort per second than its synthetic counterpart, providing a potentially more physiologically meaningful assessment of collapse severity. While dynamic airway CFD simulations in RS patients have not previously been reported, similar WOB magnitudes were observed by Gunatilaka et al. [23] in patients with tracheomalacia, where the WOB per day was reported to be 437.2 ± 454.1 J, similar to the collapsing case shown here, with a WOB value of 652.3 J/day (7.55 mJ/s * 86400 s/day). TABLE 1Patient data and airflow metrics.CasesNon‐collapsingSyntheticCollapsingWeight (kg)4.43.073.07Breathing cycle time (s)1.11.01.0Tidal volume (ml)20.1712.812.8Peak inspiration velocity (m/s)11.822.839.2Peak expiration velocity (m/s)12.420.820.8Peak inspiration inlet/lung pressure (Pa)−119−267−713Peak expiration inlet/lung pressure (Pa)137237237WOB/(day*weight) (J/kg)62.5103.6212.2WOB/cycle (mJ)3.503.687.55WOB/cycle inspiration (mJ)1.712.015.88WOB/cycle expiration (mJ)1.791.671.67 Patient data and airflow metrics. Evans AK et al. [37] state “RS describes a group of physical findings that are not only causally heterogeneous, but pathogenetically and phenotypically variable”. According to the current RS clinical consensus [38], glossoptosis is the primary cause of respiratory stress in RS. It's a dynamic clinical condition characterized by the displacement of the tongue base into the oropharynx and laryngopharynx, which aligns with our findings of increased work of breathing in these regions. The simulations of the full respiratory cycle highlight how airway collapse exacerbates energy losses through jet formation and abrupt flow cut‐off. Prior to collapse, flow in the collapsing region of the upper airways develops substantially higher inspiratory velocities to match the same tidal volume over a shortened inspiration time, producing a high‐speed jet that persists downstream as it propagates into the lower oropharynx and laryngopharynx. This high‐speed jet generates strong shear layers, associated to high viscous and turbulent dissipation. As collapse initiates, flow is rapidly decelerated at the constriction, producing a localized dissipation spike in the upper oropharynx, followed by near‐zero dissipation once flow is fully cut off. Region‐specific (WOB) quantifies these mechanisms: the lower oropharynx dominates inspiratory losses in both cases due to downstream jet–wall interactions, while the collapsing case exhibits an additional dissipation peak associated with the onset of collapse. These observations demonstrate that airway collapse is inherently a time‐dependent phenomenon, and that comparing temporally integrated metrics between collapsing and non‐collapsing scenarios is important for providing a quantitative and mechanistic understanding of the collapse process and its impact on breathing dynamics in patient‐specific cases. Limitations The magnitude of the inlet flow rate is determined using an empirical relation between body weight and tidal volume in healthy infants, which may not fully capture the altered respiratory mechanics of RS patients and, thus, introduces some uncertainty in the absolute flow magnitude. However, this approach provides a useful benchmark of the respiratory effort required for an RS infant to breathe at the same level as a healthy counterpart of the same weight. Clinically, patients often generate increased, dynamically varying lung pressures to meet normal metabolic oxygen demands when airway obstruction occurs. By adopting this approach as a common baseline, we can still consistently characterize and compare the severity of airway obstruction across cases. The method presented here uses an idealized mass flow rate waveform temporally aligned and scaled using chest movement data, thereby improving on our previous approach to time the respiratory cycle with the dynamic 4D‐CT imaging, which relied solely on airway volume changes [25]. Airway volume alone may be misleading because neuromuscular motion of the airway wall can misrepresent the true respiratory phase, the wall may deform actively in response to pressure loading rather than passively contract or expand with inspiration and expiration, respectively. However, this approach still has limitations, as airway collapse can introduce phase shifts between thoracic motion and airflow, notably, our data show continued chest expansion even during periods of airway collapse (Figure 1). A slight shift in the waveform could impact the results, so we are exploring integrating a pressure‐cannula measurement during 4DCT acquisition, which would provide a more physiologically accurate representation of the respiratory waveform, overcoming this limitation. A final limitation is the finite resolution of the CT images. Because we use a time‐dependent resistance at the collapsing airway region, rather than closing the cross‐sectional area at the region where collapse applies (for which we have no time‐resolved imaging nor could be done without geometrical singularities and computational instabilities associated with them), we assume a harmonic rate of reduction in airway radius to compute the flow resistance during collapse. This simplification directly influences the rate at which collapse accelerates and then slows down the flow through the constriction, as it closes, and consequently governs the dissipation spikes observed in the upper oropharynx. The timing and duration of collapse are also critical, as they determine how the inspiratory waveform is scaled to match the tidal volume of the non‐collapsing equivalent (synthetic) case. For instance, a slight shift in collapse timing from peak inspiration to mid‐inspiration reduces the scaled mass flow rate, since the collapse occurs when the flow is lower and less volume is added during the remaining inspiration waveform, leading to an underestimation of flow metrics. Conversely, a longer collapse duration shortens the effective inspiration time, increasing the scaled mass flow rate and potentially leading to an overestimation of flow metrics. Conclusion This study introduces the first patient‐specific CFD framework that integrates 4DCT‐derived airway wall motion with a representation of airway collapse, to quantify airflow dynamics and respiratory effort across the full breathing cycle in infants with RS. By explicitly modeling time dependent airway motion and collapse, the simulations capture the flow dynamics that are governed by the collapse and that cannot be resolved using static geometries or peak‐only metrics. Comparison with a synthetic non‐collapsing equivalent case demonstrates that airway collapse markedly increases inspiratory velocities, pressures, and localized energy dissipation. Overall, these results establish a dynamic CFD method to assess airway collapse severity in RS, providing a foundation for more patient‐specific quantitative evaluations of upper airway obstruction. The study describes a novel method and highlights its feasibility for characterizing the severity of RS patients with upper airway obstruction, especially those including collapse. This method has been described in detail and demonstrated on collapsing and non‐collapsing cases. With this foundation, it can now be applied to a larger cohort of patients and a correlation between engineering and clinical metrics can be developed so that it can be used by clinicians to inform surgical decisions. Funding National Institutes of Health, Grant/Award Number: 5R21DE033120, grant/award number: 5R21HL172011. Ethics Statement The patients signed informed consent forms and were recruited with a protocol approved by the Seattle Children's Hospital IRB. Conflicts of Interest The authors declare no conflicts of interest. Supporting information Data S1: Effect of mesh resolution and numerical solver settings on energy imbalance.