Diffusion MRI ‐Based Fetal Thermometry Using Cerebrospinal Fluid: A Translational Feasibility Study

ABSTRACT Background Noninvasive fetal temperature assessment could improve MRI safety monitoring and provide insight into fetal physiology. Diffusion MRI is a potential approach because water diffusivity in free fluids is temperature dependent. However, the applicability of adult‐derived diffusivity–temperature calibration to neonatal and fetal CSF remains uncertain. Purpose To evaluate the feasibility of diffusion MRI for noninvasive estimation of a diffusion‐derived proxy of fetal temperature in cerebrospinal fluid (CSF). Study Type Prospective feasibility study. Population/Subjects 11 preterm neonates (postmenstrual age, 30.0–40.6 weeks; 5 male, 6 female) underwent CSF diffusion MRI for translational validation, and 11 fetuses (gestational age, 27.6–37.6 weeks) underwent CSF diffusion MRI for in utero feasibility assessment. Field Strength/Sequence 1.5T monopolar spin‐echo echo‐planar diffusion imaging for neonatal validation; 3T monopolar spin‐echo echo‐planar diffusion MRI for fetal imaging. Assessment Temperature was derived from CSF apparent diffusion coefficients using an established diffusivity–temperature calibration. Neonatal estimates were compared with pre and postscan axillary temperatures. Monte Carlo simulations evaluated the effects of noise, gradient‐amplitude scaling errors, and calibration uncertainty. Statistical Tests Bland–Altman analysis, linear regression, descriptive statistics, Huber M‐estimation, and Monte Carlo simulations were performed. Statistical significance for regression analyses was defined as p < 0.05. Results Diffusion‐derived CSF temperatures agreed with axillary measurements within approximately ±1°C in 11 preterm neonates and showed Bland–Altman biases of 0.41°C and 0.50°C for mean diffusion‐derived temperature compared with pre and postscan axillary measurements, respectively. In 11 fetuses, the CSF temperature estimates using the Huber M‐estimator were 37.13°C ± 0.41°C (mean ± SD). Monte Carlo simulations revealed a decrease in the temperature SD from about 4.5°C at SNR = 30 to 1.6°C at SNR = 80. Gradient‐amplitude scaling errors of ±8% resulted in approximately 6.5°C–7°C absolute temperature bias, which was reduced to ≤ 0.1°C after ideal effective b value correction. Data Conclusion Diffusion MRI enables feasible estimation of a diffusion‐derived proxy of fetal temperature in CSF. Level of Evidence 2. Technical Efficacy Stage 1.