da Costa Campos, L. | bioRxiv Science

bioRxiv2022

Towards robust and generalizable super-resolution generative adversarial networks for magnetic resonance neuroimaging: a cross-population approach

Magnetic resonance imaging (MRI) is fundamental to neuroscience, where detailed structural brain scans improve clinical diagnoses and provide accurate neuroanatomical information. Apart from time-consuming scanning protocols, higher image resolution can be obtained with super resolution algorithms. We investigated the generalization abilities of Super Resolution Generative Adversarial Neural Networks (SRGANs) across different populations. T1-weighted scans from three large cohorts were used, spanning older subjects, newborns, and patients with brain tumor- or treatment-induced tissue changes. Upsampling quality was validated using synthetic and anatomical metrics. Models were first trained on each cohort, yielding high image quality and anatomical fidelity. When applied across cohorts, no artifacts were introduced by the SRGANs. SRGANs that were trained on a dataset combining all cohorts also did not induce any population-based artifacts. We showed that SRGANs provide a prime example of robust AI, where application on unseen populations did not introduce artifacts due to training data bias (e.g., insertion or removal of tumor-related signals and contrast inversion). This is an important step in the deployment of SRGANs in real-world settings.

neuroscience↗

bioRxiv2020

The role of thickness inhomogeneities in hierarchical cortical folding

The morphology of the mammalian brain cortex is highly folded. For long it has been known that specific patterns of folding are necessary for an optimally functioning brain. On the extremes, lissencephaly, a lack of folds in humans, and polymicrogyria, an overly folded brain, can lead to severe mental retardation, short life expectancy, epileptic seizures, and tetraplegia. The construction of a quantitative model on how and why these folds appear during the development of the brain is the first step in understanding the cause of these conditions. In recent years, there have been various attempts to understand and model the mechanisms of brain folding. Previous works have shown that mechanical instabilities play a crucial role in the formation of brain folds, and that the geometry of the fetal brain is one of the main factors in dictating the folding characteristics. However, modeling higher-order folding, one of the main characteristics of the highly gyrencephalic brain, has not been fully tackled. The effects of thickness inhomogeneity in the gyrogenesis of the mammalian brain are studied in silico. Finite-element simulations of rectangular slabs are performed. The slabs are divided into two distinct regions, where the outer layer mimics the gray matter, and the inner layer the underlying white matter. Differential growth is introduced by growing the top layer tangentially, while keeping the underlying layer untouched. The brain tissue is modeled as a neo-Hookean hyperelastic material. Simulations are performed with both, homogeneous and inhomogeneous cortical thickness. The homogeneous cortex is shown to fold into a single wavelength, as is common for bilayered materials, while the inhomogeneous cortex folds into more complex conformations. In the early stages of development of the inhomogeneous cortex, structures reminiscent of the deep sulci in the brain are obtained. As the cortex continues to develop, secondary undulations, which are shallower and more variable than the structures obtained in earlier gyrification stage emerge, reproducing well-known characteristics of higher-order folding in the mammalian, and particularly the human, brain.

neuroscience↗