Development of a composite diffusion tensor imaging score correlating with short-term neurological status in neonatal hypoxic-ischemic encephalopathy

Abstract

Hypoxic-ischemic encephalopathy (HIE) is the most common cause of neonatal acquired brain injury. Although conventional MRI may predict neurodevelopmental outcomes, accurate prognostication remains difficult. As diffusion tensor imaging (DTI) may provide an additional diagnostic and prognostic value over conventional MRI, we aimed to develop a composite DTI (cDTI) score to relate to short-term neurological function. Sixty prospective neonates treated with therapeutic hypothermia (TH) for HIE were evaluated with DTI, with a voxel size of 1 × 1 × 2 mm. Fractional anisotropy (FA) and mean diffusivity (MD) from 100 neuroanatomical regions (FA/MD ^*100 = 200 DTI parameters in total) were quantified using an atlas-based image parcellation technique. A least absolute shrinkage and selection operator (LASSO) regression was applied to the DTI parameters to generate the cDTI score. Time to full oral nutrition [short-term oral feeding (STO) score] was used as a measure of short-term neurological function and was correlated with extracted DTI features. Seventeen DTI parameters were selected with LASSO and built into the final unbiased regression model. The selected factors included FA or MD values of the limbic structures, the corticospinal tract, and the frontotemporal cortices. While the cDTI score strongly correlated with the STO score (rho = 0.83, p = 2.8 × 10^-16), it only weakly correlated with the Sarnat score (rho = 0.27, p = 0.035) and moderately with the NICHD-NRN neuroimaging score (rho = 0.43, p = 6.6 × 10^-04). In contrast to the cDTI score, the NICHD-NRN score only moderately correlated with the STO score (rho = 0.37, p = 0.0037). Using a mixed-model analysis, interleukin-10 at admission to the NICU (p = 1.5 × 10^-13) and tau protein at the end of TH/rewarming (p = 0.036) and after rewarming (p = 0.0015) were significantly associated with higher cDTI scores, suggesting that high cDTI scores were related to the intensity of the early inflammatory response and the severity of neuronal impairment after TH. In conclusion, a data-driven unbiased approach was applied to identify anatomical structures associated with some aspects of neurological function of HIE neonates after cooling and to build a cDTI score, which was correlated with the severity of short-term neurological functions.

Keywords: diffusion tensor imaging; hypoxic-ischemic encephalopathy; least absolute shrinkage and selection operator; neonatal brain atlas; outcome prediction; serum biomarkers; short-term neurologic outcome.

Figure 1

Demographic histograms of 60 subjects with neonatal HIE who underwent therapeutic hypothermia treatment. Categorical clinical variables are summarized in (A–F) with color scaled by variables, and continuous clinical variables are summarized in (G–J).

Figure 2

Representative parcellation maps superimposed on FA maps. The locations and laterality (L/R) of the selected 17 structures in the cDTI score calculation are annotated. (A–D) Axial images at the level of the corticospinal tract, the uncinate fasciculus (A), the cerebral peduncle (B), the basal ganglia (C), and the superior parietal lobule (D). (E) A sagittal image at the level of the right fornix and the right cuneus.

Figure 3

Scatterplots showing the relationship among severity scales: comparison between (A) the cDTI score and the STO score, (B) the NICHD-NRN score and the STO score, and (C) the NICHD-NRN score and the cDTI score. Solid black lines with gray areas represent the regression lines with 95% confidence intervals, and Spearman's correlation coefficients/p-values are shown in the upper left corner of each graph. For (B), the data are jittered to show the sample size.

Figure 4

(A) Spearman's correlation coefficient matrix between the cDTI score and categorical clinical variables (i.e., STO, Sarnat, NICHD-NRN, sex, field strength, scanner preference). (B) Pearson's correlation coefficient matrix between the cDTI scores and numerical variables (postmenstrual age, chronological age, gestational age, and weight). Color-coded numbers in the upper right half of the matrix indicate correlation coefficients (*p < 0.05, **p < 0.01, and ***p < 0.001, blue: positive coefficient, red: negative coefficient). The color-coded ellipses in the lower left half of the matrix indicate the strength of correlation between variables, with blue indicating a negative correlation and red indicating a positive correlation. The shape of the ellipses indicates the strength of the correlation (ellipses are sharp when the correlation is strong and round when it is weak), positive slope indicates a positive correlation, and negative slope indicates a negative correlation.

Figure 5

Time courses of biomarker values [(A): IL-10 and (B): tau] by severity group. Raw biomarker values are shown as scatterplots, and time courses are indicated as error bars. The error bars on each timepoint (baseline, during TH, end of TH/rewarming, and after rewarming) were calculated based on the results of the mixed-model analysis. Significance stars are embedded according to the result of the post-hoc t-test (*p < 0.05, **p < 0.01, and ***p < 0.001).

Figure 6

Scatterplots illustrating the relationship between the cDTI score and biomarker values [(A): IL-10 and (B): tau] over time (baseline, during TH, end of TH/rewarming, and after rewarming). Solid lines with gray areas indicate the regression lines with 95% confidence intervals, and Spearman's correlation coefficients/p-values are shown in the upper left corner of each graph.