Subjects
The study was conducted in the MRI suite contained within the neonatal intensive care unit (NICU) at the Alexandra Cohen Hospital for Women and Newborns at New York-Presbyterian. Twelve infants were imaged in the study (4 Males/8 Females), median PMA at study 40.0 wks, IQR(38.0 wks,49.0 wks), median gestational age (GA) 29.1 wks, IQR(26.7 wks,30.9 wks), median weight at scan 3.1 kg, IQR (2.6 kg, 3.8 kg), median weight at birth 0.88 kg, IQR(0.65 kg, 1.21 kg), median height 48.8 cm, IQR(45.5 cm, 51.5 cm) (Table 1). Participants were enrolled if a clinically indicated MRI of the brain was scheduled. One additional subject was enrolled and scanned but excluded due to rotational motion during the PREFUL MRI scan. Additional research imaging of the lung was performed following a clinically prescribed brain scan during the period May 2021 through March 2022. A separate protocol approved by the Institutional Review Board at Weill Cornell Medicine with informed consent was obtained from parents of all subjects for MR imaging of the lung.
Magnetic Resonance Imaging Preparation and Clinical Monitoring
Infants were prepared for the MRI examination using a “feed and swaddle” technique by nursing staff in the NICU. Infants were fed and allowed to fall into a natural sleep without sedation or anesthesia. They were provided three layers of hearing protection (soft foam earplugs, plastic ear muffs, and blanket padding) and securely swaddled with blankets. A pulse oximeter was placed on the foot and the infant’s heart rate and oxygen saturation were monitored through the duration of the scan. Signals of an infant waking from sleep, such as loss of O2 saturation tracing or persistent increase in heart rate, prompted examiners to enter the MRI room to check on the infant immediately. The scan was stopped if the infant did not quickly fall back to sleep [15].
Magnetic Resonance Imaging Acquisition
All imaging was performed on a 1.5 Tesla (T) MRI scanner (MAGNETOM Amira; Siemens Healthcare, Erlangen Germany) located within the NICU using a pair of 8-channel flex coils with a coverage of 20 cm x 22 cm (Noras MRI product GmbH; Höchberg, Germany). A single coronal slice at the level of the trachea using 2D spoiled gradient echo (TFL) sequence was acquired with a 168 ms repetition time (TR), 1345 Hz/Pixel receive bandwidth, 1.08 ms echo time (TE), 5⁰ flip angle (FA), 10 mm slice thickness, 25 cm field of view (FOV) and a 128x102 matrix reconstructed to 256 x 256 resulting in an in-plane resolution of 1 mm x 1 mm. Four hundred images were acquired continuously over 67 seconds without respiratory or cardiac gating. Two additional slices were placed anterior and posterior respectively to the central PREFUL slice and individually acquired under separate series. A multiplicative reconstruction factor of 25 was applied to get a full range of signal intensities in the lung.
Magnetic Resonance Imaging Quantitative Analysis
Image analysis was performed in MATLAB using the PREFUL analysis routine developed by several of the authors (AV, JVC) [16]. In addition, an interactive prototype application, MRLung 2.0 (RG, Siemens Healthcare; Erlangen, Germany) was used for rapid visualization of the data. Phase-resolved functional lung (PREFUL) imaging is a method to evaluate dynamic MRI data acquired in free-breathing without the need for contrast agent administration. Artificial intelligence (AI) segmentation of the lungs was performed with the ability to manually edit the contour to include additional regions in the periphery [Figure 1]. The acquired image time series was transformed to a fixed lung volume with a group-oriented registration approach using code developed in MATLAB (AV, J-VC) [17]. This enabled the analysis of the signal time course in each voxel, which was mainly influenced by two variables.
First, the proton density (PD) of lung tissue is reduced compared to normal soft tissues such as muscle and fat. This decrease in PD translates to a reduction in the amount of MRI signal obtained from each voxel compared to other tissues. The MR signal intensity also changes between inspiration and expiration. The amount of hydrogen in normal air is essentially zero (~0.6 ppm) and not visible with MR imaging at 1.5 T. During inspiration, the lungs fill with air and increase in volume reducing the amount of lung tissue in each voxel resulting in a signal decrease compared with expiration. There is also an 8 ppm bulk magnetic susceptibility difference between air and lung tissue (alveoli). This effectively shortens the T2* of the voxels in the lungs requiring a TE as short as possible (≈ 1 ms) to maximize the MR signal intensity before decay occurs. A high receive bandwidth is also desirable to minimize the dephasing in the region. Additionally, there is a time-of-flight (TOF) blood flow effect, which describes the inflow of fresh spins with a higher magnetization into the imaging plane such as in MR angiography. Signal will increase and decrease in the time course curve in accord with the cardiac cycle. While the first effect is directly connected to ventilation, the second is related to perfusion. Since these signal changes occur at different frequencies (cardiac frequency vs respiration frequency), they can be separated with an adequate filtering approach. Specifically, a low-pass filter was applied to evaluate ventilation and a high-pass filter was used for perfusion, respectively. Further, an image-guided, edge-preserving filter was applied to increase the signal-to-noise ratio while preserving fine structural details [18].
In the next step, the respiratory and cardiac phase were estimated for each image with a sine-model. The sorting of the images allows one to interpolate a complete cardiac and respiratory cycle. This 3D ventilation (V) and perfusion (Q) information (2 spatial dimensions + 1 temporal dimension) can be further processed to calculate a series of V/Q parameters, including regional ventilation, quantified perfusion, and flow-volume correlation metric [19,20]. By applying thresholds, binary parameters can be derived and combined to produce V/Q maps, which can provide the percentage of lung volume affected by ventilation and perfusion defects and the respective match and mismatch ratios.[Figure 2] The following parameters are reported as results: ventilation defect percentage (VDP%), perfusion defect percentage (QDP%), V/Q defect and non-defect match percentage (VQM%). The ventilation defect percentage (VDP%) is the percent of the lung in that slice that is below the median ventilation – 4 standard deviations [15]. Similarly, the perfusion defect percentage (QDP%) is the percent of lung in that slice that is below the median perfusion – 1 standard deviation. The VDP% and QDP% are reported as both total and exclusive with exclusive indicating no overlap of defect in V and Q maps. The V/Q defect maps show the match percentage between V and Q defects and non-defects in that slice.
Statistical Analysis
Descriptive statistics were computed for all quantitative measures by reporting the median and interquartile range. A two-tailed Mann-Whitney or Wilcoxon Rank Sum non-parametric t-test was performed for each of the ventilation and perfusion parameters between groups of BPD Grade 0/1 (n=6) and BPD Grade 2/3 (n=6) with significance defined as p<0.05. The non-parametric test was chosen given normality could not be assumed with this sample size. An online toolbox was used to calculate the median, IQR and box plots.