49 Mireia Alenyà Computer Vision News The human brain develops from a smooth cortical surface in early stages of fetal life to a convoluted one postnatally, creating an organized ensemble of folds. This is clinically relevant since abnormal folding patterns are linked to neurodevelopmental disorders such as autism, schizophrenia, or epilepsy. Recent advancements in MRI technology, that allow to collect fetal data at these early stages of pregnancy, have enabled researchers like Mireia to explore how essential brain folds form during gestation. During her PhD, Mireia and colleagues constructed a 3D patient-specific computational pipeline for analysing fetal brain development (see figure in the left page), beginning with personalised fetal and neonatal MRI scans. The pipeline consisted of segmentation of the grey matter tissue, construction and processing of 3D meshes, and incorporated a Finite Element Method (FEM) model that allowed for virtual simulations of brain growth and folding (see figure below), providing insights into developmental trajectories and potential outcomes. Additionally, several evaluation metrics were developed to quantitatively compare these simulations with real brain development data collected over time. However, the computational complexity of this pipeline, including significant computational time requirements for some stages, led Mireia to the second part of her research. Here, she focused on accelerating these computationally expensive aspects of the pipeline using deep learning methods. This involved two main components: Firstly, Mireia participated in the Fetal Tissue Annotation Challenge (FeTA) (2021 & 2022) organised within the MICCAI conference, to develop algorithms that automatically segmented fetal brain tissues within MRI scans. These algorithms were designed to adapt to various data types and even included techniques to artificially expand the datasets used for training (data augmentation). Secondly, Mireia developed a "surrogate" deep learning model to accelerate complex FEM simulations. This model acted as a shortcut, learning to predict the outcomes of FEM simulations in a fraction of the time, significantly reducing computation time from 48 hours to a few seconds. Overall, Mireia's groundbreaking work represented a significant leap forward in understanding fetal brain development. Simulation example of brain growth and folding.
RkJQdWJsaXNoZXIy NTc3NzU=