Automatic Fetal Brain Surface Extraction for Analysing the Neural Basis of Embodiment

Abstract

Introduction

Behavioral research in fetuses suggests that functional sensory connections in the human brain already develop during gestation. Dubois et al. [1] suggest using changes in brain anatomy as biomarkers for functional and cognitive development. Fetuses show physical and behavioral reactions to sounds, touch, light, and taste in different experiments as early as 24 gestational weeks (GW) [2].

Theories from embodied cognition deem this early experience of sensory feedback from the environment crucial. It is necessary for later integration of different senses and development of higher-order cognitive functions. It is suggested that the interaction of the fetus with intrauterine stimuli has a direct influence on the development of the fetal brain [2].

Challenges and Problem Statement

The fetal brain can be imaged by fast-sequence fetal brain Magnetic Resonance Imaging (MRI). Current approaches for morphometric analysis of fetal brain development rely on traditional atlas-based methods for fetal brain surface extraction [2] in combination with cortical surface parcellation [3]. Those are limited in their robustness to variations in relative positions of the fetal brains, and a precise prior registration of the cortical surface is necessary [2]. Further fetal brain MRI challenges include imaging artifacts due to fetal motion, intensity differences due to myelination, and changing shape and size due to cortical folding.

Motivation and Aim

Investigation of the longitudinal morphological development of the cortex in the fetal brain, through cortical surface extraction, can lead to new insights into the neural basis of early embodied cognition and inform theories of cognitive development [2]. The aim is to evaluate progressive changes in surface-based features and contextualize them within fetal cognitive development literature.  Further, advantages as well as disadvantages of deep learning based fetal brain surface extraction methods compared to traditional methods as a baseline will be evaluated, as a basis for the cortical feature extraction. 

Method and Experimental Setup

A processing workflow (pipeline) for automatic fetal brain surface extraction based on fetal brain MRI scans will be developed. The pipeline will consist of high-resolution reconstruction, surface extraction, and cortical feature extraction. Different deep learning based techniques will be compared and evaluated on a data set of around 200 high-resolution reconstructed neurotypical T2-weighted fetal brain MRI scans ranging from 17 to 38 GW. Based on the extracted cortical surface, the morphological development of the cortex during gestation will be investigated through features like gyrification index or sulcal depth, and discussed in relation to behavioral data available in the literature, as no behavioral data are available for the fetuses in the data set. The discussion will be done from an embodied cognition viewpoint, to show that embodied actions are critical for fetal brain development [2].

References 

[1] J. Dubois, G. Dehaena-Lambertz, S. Kulikova, C. Poupon, P. S. Hüppi, and L. Hertz-Pannier, “The early development of brain white matter: A review of imaging studies in fetuses, newborns and infants,” Neuroscience, vol. 276, pp. 48-71, 2014.

[2] L. Craighero, “An embodied approach to fetal and newborn perceptual and sensorimotor development,” Brain and Cognition, vol. 179, Article 106184, 2024.