Neuroscience

BOUNTI: Brain vOlumetry and aUtomated parcellatioN for 3D feTal MRI

School of Imaging Sciences and Biomedical Engineering, King’s College London, London, UK
Centre for the Developing Brain, King’s College London, London, UK
Department of Computing, Imperial College London, London, UK
Biomedical Image Technologies, ETSI Telecomunicacion, Universidad Politécnica de Madrid and CIBER-BBN, ISCII, Madrid, Spain
Department of Congenital Heart Disease, Evelina London Children’s Hospital, London, UK

https://doi.org/10.7554/eLife.88818.1

Open access
Copyright information

Figures and data

The main steps of the proposed solution for implementation of a deep learning pipeline for 3D brain tissue parcellation for motioncorrected 3D fetal MRI.

Proposed BOUNTI pipeline for automated brain tissue parcellation for 3D T2w fetal MRI (A) and preparation of labels for semi-supervised training (B).

Summary information about training and testing datasets.

Proposed tissue parcellation protocol defined in the dHCP fetal brain MRI atlas.

Proposed multi-tissue brain parcellation maps at 21, 26, 31 and 36 weeks GA timepoints of the dHCP fetal brain MRI atlas.

An example of comparison of BOUNTI segmentation results vs. the original and manually refined (ground truth) label propagation: late GA test case from dHCP cohort (3T, TE=250ms).

Examples of the outputs of segmentation pipeline for the test cases with different acquisition protocols and gestational ages.

Examples of BOUNTI segmentation results for suboptimal image quality (A) and abnormal Payette et al. (2021) (B) cases.

Examples of visual comparison of BOUNTI results with classical 3D UNet trained on the original FETA Payette et al. (2021) datasets with manual labels (7 ROIs) and label propagation from an alternative atlasGholipour et al. (2017) (27 ROIs): two datasets at 24 and 35 weeks GA are from 3T, TE=180ms cohort. Note: the label colours were adapted for optimal comparison.

Quantitative comparison of BOUNTI segmentation results with semi-manual (label propagation with manual refinement) ground truth labels: average Dice, recall and precision. The datasets include 40 cases from 21-38 weeks GA range from 4 different acquisition protocols. Note: the labels for pair/related structures were combined.

Growth charts: volumetry of the major brain structures for normal fetal cohorts from fetal MRI research studies with different acquisition parameters: 1.5T, TE=180ms (blue, 55 participants); 3T, TE=180ms (red, 91 participants); 3T, TE=250ms (lilac, 244 participants). Note: the left and right labels for pair structures were combined.

Comparison of the VM (65) and normal control (60) fetal MRI cases from Kyriakopoulou et al. (2014) study based on the new set of unedited parcellations from BOUNTI: lateral ventricle and supratentorial brain volumes. Note: the left and right labels for pair structures were combined.

Sign up for email alerts