Real‐time fetal brain tracking for functional fetal MRI

Abstract

PurposeTo improve motion robustness of functional fetal MRI scans by developing an intrinsic real‐time motion correction method. MRI provides an ideal tool to characterize fetal brain development and growth. It is, however, a relatively slow imaging technique and therefore extremely susceptible to subject motion, particularly in functional MRI experiments acquiring multiple Echo‐Planar‐Imaging‐based repetitions, for example, diffusion MRI or blood‐oxygen‐level‐dependency MRI.MethodsA 3D UNet was trained on 125 fetal datasets to track the fetal brain position in each repetition of the scan in real time. This tracking, inserted into a Gadgetron pipeline on a clinical scanner, allows updating the position of the field of view in a modified echo‐planar imaging sequence. The method was evaluated in real‐time in controlled‐motion phantom experiments and ten fetal MR studies (17 + 4‐34 + 3 gestational weeks) at 3T. The localization network was additionally tested retrospectively on 29 low‐field (0.55T) datasets.ResultsOur method achieved real‐time fetal head tracking and prospective correction of the acquisition geometry. Localization performance achieved Dice scores of 84.4% and 82.3%, respectively for both the unseen 1.5T/3T and 0.55T fetal data, with values higher for cephalic fetuses and increasing with gestational age.ConclusionsOur technique was able to follow the fetal brain even for fetuses under 18 weeks GA in real‐time at 3T and was successfully applied “offline” to new cohorts on 0.55T. Next, it will be deployed to other modalities such as fetal diffusion MRI and to cohorts of pregnant participants diagnosed with pregnancy complications, for example, pre‐eclampsia and congenital heart disease.

Scanner-based real-time 3D brain+body slice-to-volume reconstruction for T2-weighted 0.55T low field fetal MRI

“…One of the possible solutions could be based on more advanced integration of the reconstruction code into the Gadgetron interface: e.g. dynamic masking, reorientation 21 and sequential addition of stacks to the reconstruction function.…”

“…Several fetal MRI research works reported scanner-based solutions for real-time brain tracking 21 and automated detection and re-acquisition of low-quality slices 22 . Both solutions implemented interaction between the scanner and external GPU-accelerated research reconstruction workstations.…”

“…This work integrates a 3D brain+body D/SVR reconstruction pipeline for T2w structural low field 0.55T fetal MRI integrated directly into the scanner environment and makes the final 3D reconstructions available during the ongoing fetal scan. It combines our previous works on automated 3D SVR reconstruction 13,32 with a real-time integrated scanner workflow 21,33 . In the context of this work, real-time integration refers to the triggering of the SVR reconstruction process immediately upon the acquisition of the last T2w stack, with retrieval of the resulting volume in the duration of the scan.…”

Fully automated planning for anatomical fetal brain MRI on 0.55T

“…A bounding box encompassing the fetal brain is calculated and used for the following landmark detection task. 18…”

“…Specifically in fetal MRI, work showed the ability to perform quality control, 14 automatic segmentation, 15,16 and automatic tracking. 17,18 Automatic field-of-view prescription was shown in the abdomen using deep learning segmentations 19 and in the heart using tracking based on landmarks 20 with successful detection ratings of 99.7%-100% for cine images and Euclidean distances between manual and automatically detected labels from 2 to 3.5 mm. Similar work in the brain detecting landmarks such as the anterior and posterior commissures and subsequently the symmetry line using multitask deep neural networks were demonstrated among others by Yang et al 21 Specifically for fetal MRI, segmentation of the eye region and detection of the general head position was suggested by Hoffmann et al 22 using classical image processing methods such as maximally stable extremal regions and by Xu et al 23 using convolutional networks-both applicable for real-time slice planning in the future.…”