Dynamic multiscale boundary conditions for 4D CT of healthy and emphysematous rats

Abstract

Changes in the shape of the lung during breathing determine the movement of airways and alveoli, and thus impact airflow dynamics. Modeling airflow dynamics in health and disease is a key goal for predictive multiscale models of respiration. Past efforts to model changes in lung shape during breathing have measured shape at multiple breath-holds. However, breath-holds do not capture hysteretic differences between inspiration and expiration resulting from the additional energy required for inspiration. Alternatively, imaging dynamically--without breath-holds--allows measurement of hysteretic differences. In this study, we acquire multiple micro-CT images per breath (4DCT) in live rats, and from these images we develop, for the first time, dynamic volume maps. These maps show changes in local volume across the entire lung throughout the breathing cycle and accurately predict the global pressure-volume (PV) hysteresis. Male Sprague-Dawley rats were given either a full- or partial-lung dose of elastase or saline as a control. After three weeks, 4DCT images of the mechanically ventilated rats under anesthesia were acquired dynamically over the breathing cycle (11 time points, ≤100 ms temporal resolution, 8 cmH2O peak pressure). Non-rigid image registration was applied to determine the deformation gradient--a numerical description of changes to lung shape--at each time point. The registration accuracy was evaluated by landmark identification. Of 67 landmarks, one was determined misregistered by all three observers, and 11 were determined misregistered by two observers. Volume change maps were calculated on a voxel-by-voxel basis at all time points using both the Jacobian of the deformation gradient and the inhaled air fraction. The calculated lung PV hysteresis agrees with pressure-volume curves measured by the ventilator. Volume maps in diseased rats show increased compliance and ventilation heterogeneity. Future predictive multiscale models of rodent respiration may leverage such volume maps as boundary conditions.

Figure 1. Block diagram illustrating the data processing flow.

Figure 2. Examples of image processing for registration testing.

A) Original image. B) Image with bones removed. C) Image with all background masked. D) Same as C, with Gaussian filter applied prior to masking. E) Same as D, with contrast enhancement applied.

Figure 3. An example of finite element discretization of an image.

Figure 4. Average landmark displacement (in pixels, red) and the percentage of misregistered landmarks (blue) for the five cases shown in Figure 1.

The red dashed line indicates the observer’s average “click error”.

Figure 5. Sample registration results.

Column A: control. Column B: full-lung dosed. Column C: single-lobe dosed. Row 1: coronal images, unprocessed. Row 2: Jacobian maps, between the lowest and highest inflation levels. Row 3: volume maps. The color scale represents the volume change (×10⁻⁶ mL) in each voxel between 0 cmH₂O and 8 cm H₂O.

Figure 6. Comparison of the average ventilator-measured volume and the total volume from the volume maps for six untreated rats, plotted as a function of tracheal pressure.

Error bars represent the standard deviation of the mean ventilator volume and tracheal pressure over the entire imaging experiment.

Figure 7. Representative pressure-volume curves made from the total volume from the volume maps and average tracheal pressure.

Data are from three different rats, one from each dose group.