Computational protocol: In Vivo Quantification of Placental Insufficiency by BOLD MRI: A Human Study

Similar protocols

Protocol publication

[…] The acquired images were first corrected for signal non-uniformity and then registered to mitigate motion with a non-rigid transformation approach designed for BOLD time series which outperforms traditional rigid body motion correction approaches (Fig. ). The bias field accounting for B1 inhomogeneities was estimated from averaged selected volumes collected in first 10 minutes by using N4ITK algorithm, and subsequently applied to correct for signal non-uniformity of each volume. For intravolume motion correction, each volume was separated into two sub-volumes: even and odd slices, then registered to each other using the group-wise approach. For intervolume motion correction, a reference volume with a least sum of mean square error (MSE) difference compared to the rest of the volumes in the series was selected. Then pairwise registration was employed between the reference volume and the other volumes. As an initialization, a six degrees of freedom rigid transformation as a mapping from the reference volume to the rest of the volumes was estimated. To compensate for motion of the deformable fetal body and placenta in the time series, a non-rigid body transformation was performed following the initial rigid body alignment. To compensate for motion of the brain, which is more rigid, a second rigid body transformation was estimated following the initial rigid body alignment. All registration steps were carried out in Elastix, an open source software. ROIs of the placenta and fetal organs in the reference frame were manually delineated using ITK-SNAP, and propagated to all time frames. Outlier volume detection was performed based on 1) over-deformation during intravolume motion correction; and 2) unexpected average signal intensity change of nearby time frames. The median fraction of outliers discarded for placenta was 3.3% (0.9–53.4%), for fetal brain was 22.1% (1.6–85%) and for fetal liver was 15.6% (1.4–34.8%). The registered 4D placenta image was subsequently smoothed in the spatial domain by a Gaussian kernel with width of 5 pixels, sigma 1.5, and interpolated in the temporal domain followed by a smoothing window of around 1 min (8–12 TRs). The indicator of oxygen saturation level change, ΔR2*, was calculated for placenta and fetal organs, as follows:1ΔR2⁎(t)=−(R2⁎−R2⁎baseline)=log(S(t)/Sbaseline)/TE=−r2⁎⋅(VF⋅Δ[dHbF(t)]+VM⋅Δ[dHbM(t)])where r2* is the relaxivity of deoxygenated hemoglobin, VF and VM are the volume fractions of fetal and maternal blood, [dHbM], [dHbF] stands for the maternal and fetal deoxygenated hemoglobin concentration repectively. When oxygenation level increases, [dHb] drops, which decreases R2* and increases BOLD signal S(t), therefore we define ΔR2* in the way that it increases with the oxygenation level increase.Using ITK-SNAP, placenta regions corresponding to each twin were determined by tracing umbilical cord that connects each fetus to its placenta. The placental tissue corresponding to each twin was manually delineated by first identifying the cord insertion for each twin and confirming with an expert radiologist. Second, as the boundary dividing portions of the placenta supplying each twin is not visible, a 5 cm boundary zone half-way between the two cords was excluded from analysis. ROIs from each cord to the respective boundary were extended circumferentially around each cord to define placental regions corresponding to each twin. The fetal brain and liver was manually delineated with boundaries supervised and confirmed by expert radiologists.Similarly to the fMRI studies in the brain, , we employ the Gamma function to capture the shape of the hemodynamic response of the placenta. Gamma function fitting was implemented both in the regions-of-interest (ROIs) and on voxel-by-voxel basis. The R2* time series in placental tissue are fit to a gamma function convolved with the oxygen paradigm using Least-Squares:2R2⁎(t|α,β,Δ,C1,C2)=C1+C2⋅(t−Δ)α−1⋅e−(t−Δ)/β⊗Poxy(t,Δ)Poxy(t,Δ)={1fort≥Δ0fort<Δ} 3τ=α⋅(β−1) 4TTP=Δ+τwhere x is the frame number after onset of signal change of the R2* time series, α and β are gamma function parameters, C1 describes baseline T2* weighted signal, and C2 describes the amplitude of R2* change; Poxy is a step function that describes administration of oxygen. With Δ we describe the delay time of oxygen arrival; together with the mode τ of the Gamma function (derived from α and β), we obtain Time-to-Plateau (TTP). Variations in fitting and TTPs are demonstrated in Table. . […]

Pipeline specifications

Software tools elastix, ITK-SNAP
Applications Magnetic resonance imaging, Functional magnetic resonance imaging
Organisms Homo sapiens
Diseases Obstetric Labor, Premature, Placental Insufficiency
Chemicals Oxygen