B1129
Title: Complex surface reconstruction and its application in three-dimensional models of the human brain
Authors: Thiago Cardoso - University College Dublin (Ireland) [presenting]
Michelle Carey - Univerity College Dublin (Ireland)
Abstract: The utilization of image data has become instrumental in driving advancements in medical research, particularly within the field of neurology. The powerful tool offers a non-invasive method to explore the intricate workings of the human brain. Recent advancements in medical software allow the generation of three-dimensional models of the human brain using Magnetic Resonance Imaging (MRI) scans. These models aim to capture better the complex geometrical features of the brain's surface, helping researchers and practitioners better understand its structure. This, in turn, can lead to more precise diagnoses and treatment strategies. However, the image acquisition process involves several steps that contaminate the data with noise, which can compromise the accuracy of subsequent medical analysis. Methods for recovering the true signal from noisy input data are investigated on the complex geometry of the brain using a functional data analysis (FDA) framework. The approach involves modelling the noisy observations through spatial regression with partial differential equation regularization, incorporating the Laplace-Beltrami operator. Furthermore, the smooth function is approximated using a neural network, resulting in a physics-informed neural network (PINN). The performance of the proposed approach is compared to standard methods found in the literature.
