This study presents fully automatic quantification of placental function using T2* MR imaging across all MR field strengths used for fetal MRI. A cross-field network for automatic placenta segmentation was trained and is complemented with several quantitative measures - providing the first comparative placental T2* MRI assessment across field strengths and allowing comparison of normal placental T2* varies across different field strengths. We demonstrate similar changes in placental T2* values, uniformity, thickness and histogram values across all field strengths. No correlation with s BMI, maternal age and placental location was detected at any field strength.
Automatic segmentation of the placenta in MRI is key for accurate and efficient analysis, especially when dealing with large datasets. This is even more important for quantitative techniques, relying on analysing data acquired at different contrast settings together. However, the constraints of the applied contrast setting often reduced the quality of these data sets and thus influences on the required segmentation. Examples are single-shot gradient echo EPI for T2* fitting or diffusion-weighted single-shot spin echo EPI for ADC quantification, both with a resolution typically lower than anatomical T2-weighted data sets and quickly changing. contrast between neighbouring tissues. Recent publications on such quantitative data, e.g., by Abulnaga et al 16 and Pietsch et al 17 explore automatic placenta segmentation in BOLD images, with a cohort limited to one field strength and not taking into account factors like maternal position or gestational age.
The inverse relationship between T2* and field strength is well established, and has been demonstrated in other human tissues: in the adult brain mean regional and whole brain changes in a small cohort has been demonstrated 28,29; similar results have been shown in musculoskeletal tissues at clinical and high-field strengths. This study demonstrates an inverse relationship between T2* and field strength which is in line with previous studies. However, the data also clearly illustrates, that the longer T2* values at low field are particularly beneficial for placental assessment, where the larger dynamic range may allow for finer discrimination in late gestation. Furthermore, while this paper focused on establishing analysis and ranges for normal placentas, T2* was also shown to be reduced in cases affected e.g., by placental insufficiency, similarly benefiting from techniques with a larger dynamic range. The possibility for finer grained analysis, e.g. of histogram shapes was demonstrated. Clear differences regarding the artefacts are also observed between the results at different field strengths. While the data at 3T is affected by geometric distortion artefacts despite image based shimming, the data at lower field strength shows no such artefacts.
We have utilised large datasets across three field strengths, all of which confer some clinical or research advantage. Acquisition protocols were robust and optimised for all field strengths, and regions of interest were conservatively segmented - both manually for the training dataset and therefore also by the network - to prevent inaccurate mean T2* values either by incorrectly identified regions of interest or partial volume effects. Analysis of multiple parameters, including those that were anticipated to be field strength independent, improved the reliability of our results and provides confidence that measurements such as volume, uniformity and thickness can be extrapolated across datasets. The calculation of normal ranges, and as a consequence the z-scores, allows clinicians and researchers to combine data or utilise the field strength most advantageous to their work, while being able to utilise prior research pertaining to their area of interest. Examples of processed T2* maps from all field strengths are made publicly available, and the trained network is available on github (ANONYMIZED LINK).
There are some limitations to this work: firstly, the acquisition protocols vary in terms of resolution, acceleration and field of view - adapted to the individual constraints given by the field strength and/or population. The population varies according to the scanner - the mean BMI at low field was larger due to the wider bore size available. No paired experiments (in terms of the same pregnant woman having scans on all three field strengths on the same day) were included due to the considerable demand this would pose to the women. Next, scanners from two manufacturers were employed, increasing generalizability but potentially also resulting in inconsistencies due to different pre-processing steps employed on the scanner before the data are extracted. However, we have maintained the same fitting and analysis routines for all datasets for consistency. In line with most placental T2* work to date no motion correction was performed, as intra-slice motion is effectively frozen due to the multi-echo set up for each slice. Inter-slice motion may affect the data which might result in areas being included twice or not at all depending on the motion pattern, although the implications of this are unlikely to vary by field strength. Segmentations on low field are facilitated by reduced geometric distortions and hence potentially more accurate geometrically, potentially explaining the lower Dice score for 3T data. Finally, the birth weight centile was not available for a significant proportion of the datasets at 1.5T limiting the value of the here conducted statistical analysis for this specific outcome measure at this field strength.
We have demonstrated reliable changes in placental T2* measurements even if imaging is undertaken at a different field strength, but that volumetric or morphological analysis should not be expected to differ between groups. In order to aid future research, we have created standardised scores for T2* across field strengths. In particular, this work gives credence to the ongoing study of fetal MRI at low field, where we demonstrate T2* behaves similarly to higher fields, but where the advantage of the intrinsically longer T2* can be utilised; it should also give confidence in interpretation when lower fields are required for clinical reasons, such as high BMI or late gestation. Creation of z-scores also confers clinical advantage meaning that, while most units are limited to one MRI scanner field strength, they can directly utilise published data from other field strengths; furthermore, automation of placental segmentation reduces expertise required and improves efficiency of obtaining T2* results. Equivalent work determining normal T2* ranges across field strengths for other fetal organs could be undertaken. Furthermore, additional work will focus on including the entire automatic pipeline into the scanner host to allow real-time availability of the z-score for any new placenta during the fetal MRI acquisition.