Biology subjects

Bollmann, S.

Publications and source records attributed to Bollmann, S..

25 records · Page 2Linked to original sources

bioRxiv2020

Predicting brain function from anatomy using geometric deep learning

Whether it be in a single neuron or a more complex biological system like the human brain, form and function are often directly related. The functional organization of human visual cortex, for instance, is tightly coupled with the underlying anatomy with cortical shape having been shown to be a useful predictor of the retinotopic organization in early visual cortex. Although the current state-of-the-art in predicting retinotopic maps is able to account for gross individual differences, such models are unable to account for any idiosyncratic differences in the structure-function relationship from anatomical information alone due to their initial assumption of a template. Here we developed a geometric deep learning model capable of exploiting the actual structure of the cortex to learn the complex relationship between brain function and anatomy in human visual cortex such that more realistic and idiosyncratic maps could be predicted. We show that our neural network was not only able to predict the functional organization throughout the visual cortical hierarchy, but that it was also able to predict nuanced variations across individuals. Although we demonstrate its utility for modeling the relationship between structure and function in human visual cortex, our approach is flexible and well-suited for a range of other applications involving data structured in non-Euclidean spaces.

neuroscience↗

bioRxiv2019

Longitudinal Automatic Segmentation of Hippocampal Subfields (LASHiS) using Multi-Contrast MRI

The volumetric and morphometric examination of hippocampus formation subfields in a longitudinal manner using in vivo MRI could lead to more sensitive biomarkers for neuropsychiatric disorders and diseases including Alzheimers disease, as the anatomical subregions are functionally specialised. Longitudinal processing allows for increased sensitivity due to reduced confounds of inter-subject variability and higher effect-sensitivity than cross-sectional designs. We examined the performance of a new longitudinal pipeline (Longitudinal Automatic Segmentation of Hippocampus Subfields [LASHiS]) against three freely available, published approaches. LASHiS automatically segments hippocampus formation subfields by propagating labels from cross-sectionally labelled time point scans using joint-label fusion to a non-linearly realigned single subject template, where image segmentation occurs free of bias to any individual time point. Our pipeline measures tissue characteristics available in in vivo high-resolution MRI scans, at both clinical (3 Tesla) and ultra-high field strength (7 Tesla) and differs from previous longitudinal segmentation pipelines in that it leverages multi-contrast information in the segmentation process. LASHiS produces robust and reliable automatic multi-contrast segmentations of hippocampus formation subfields, as measured by higher volume similarity coefficients and Dice coefficients for test-retest reliability and robust longitudinal Bayesian Linear Mixed Effects results at 7 T, while showing sound results at 3 T. All code for this project including the automatic pipeline is available at https://github.com/CAIsr/LASHiS

neuroscience↗

bioRxiv2019

Non-Linear Realignment Improves Hippocampus Subfield Segmentation

Participant movement can deleteriously affect MR image quality. Further, for the visualization and segmentation of small anatomical structures, there is a need to improve image quality, specifically signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR), by acquiring multiple anatomical scans consecutively. We aimed to ameliorate movement artefacts and increase SNR in a high-resolution turbo spin-echo (TSE) sequence acquired thrice using non-linear realignment in order to improve segmentation consistency of the hippocampus subfields. We assessed the method in young healthy participants, Motor Neurone Disease patients, and age matched controls. Results show improved image segmentation of the hippocampus subfields when comparing template-based segmentations with individual segmentations with Dice overlaps N=51; ps < 0.001 (Friedmans test) and higher sharpness ps < 0.001 in non-linearly realigned scans as compared to linearly, and arithmetically averaged scans.

neuroscience↗

bioRxiv2019

Bayesian population receptive field modeling in human somatosensory cortex

Somatosensation is fundamental to our ability to sense our body and interact with the world. Our body is continuously sampling the environment using a variety of receptors tuned to different features, and this information is routed up to primary somatosensory cortex. Strikingly, the spatial organization of the peripheral receptors in the body are well maintained, with the resulting representation of the body in the brain being referred to as the somatosensory homunculus. Recent years have seen considerable advancements in the field of high-resolution fMRI, which have enabled an increasingly detailed examination of the organization and properties of this homunculus. Here we combined advanced imaging techniques at ultra-high field (7T) with a recently developed Bayesian population receptive field (pRF) modeling framework to examine pRF properties in primary somatosensory cortex. In each subject, vibrotactile stimulation of the fingertips (i.e., the peripheral mechanoreceptors) modulated the fMRI response along the post-central gyrus and these signals were used to estimate pRFs. We found the pRF center location estimates to be in accord with previous work as well as evidence of other properties in line with the underlying neurobiology. Specifically, as expected from the known properties of cortical magnification, we find a larger representation of the index finger compared to the other stimulated digits (middle, index, little). We also show evidence that the little finger is marked by the largest pRF sizes. The ability to estimate somatosensory pRFs in humans provides an unprecedented opportunity to examine the neural mechanisms underlying somatosensation and is critical for studying how the brain, body, and environment interact to inform perception and action.

neuroscience↗

bioRxiv2019

SHARQnet - Sophisticated Harmonic Artifact Reduction in Quantitative Susceptibility Mapping using a Deep Convolutional Neural Network

Quantitative susceptibility mapping (QSM) reveals pathological changes in widespread diseases such as Parkinsons disease, Multiple Sclerosis, or hepatic iron overload. QSM requires multiple processing steps after the acquisition of magnetic resonance imaging (MRI) phase measurements such as unwrapping, background field removal and the solution of an ill-posed field-to-source-inversion. Current techniques utilize iterative optimization procedures to solve the inversion and background field correction, which are computationally expensive and lead to suboptimal or over-regularized solutions requiring a careful choice of parameters that make a clinical application of QSM challenging. We have previously demonstrated that a deep convolutional neural network can invert the magnetic dipole kernel with a very efficient feed forward multiplication not requiring iterative optimization or the choice of regularization parameters. In this work, we extended this approach to remove background fields in QSM. The prototype method, called SHARQnet, was trained on simulated background fields and tested on 3T and 7T brain datasets. We show that SHARQnet outperforms current background field removal procedures and generalizes to a wide range of input data without requiring any parameter adjustments. In summary, we demonstrate that the solution of ill-posed problems in QSM can be achieved by learning the underlying physics causing the artifacts and removing them in an efficient and reliable manner and thereby will help to bring QSM towards clinical applications.

neuroscience↗

bioRxiv2018

DeepQSM - Using Deep Learning to Solve the Dipole Inversion for MRI Susceptibility Mapping

Quantitative susceptibility mapping (QSM) aims to extract the magnetic susceptibility of tissue from magnetic resonance imaging (MRI) phase measurements. The mapping of magnetic susceptibility in vivo has gained broad interest in several fields of science and medicine because it yields relevant information on biological tissue properties, predominantly myelin, iron and calcium. Thereby, QSM can also reveal pathological changes of these key components in devastating diseases such as Parkinsons disease, Multiple Sclerosis, or hepatic iron overload. As QSM requires the solution of an ill-posed field-to-source-inversion, current techniques utilize manual optimization of regularization parameters to balance between smoothing, artifacts and quantification accuracy. We trained a fully convolutional deep neural network - DeepQSM - to invert the magnetic dipole kernel convolution. This network is capable of solving the ill-posed field-to-source inversion on real-world in vivo MRI phase data without the need for manual parameter tuning, which proves that this network has generalized the underlying mathematical principle of the dipole inversion. We demonstrate that DeepQSMs susceptibility maps enable identification of deep brain substructures that are not visible in MRI phase data and provide information on their respective magnetic tissue properties. We illustrate DeepQSMs clinical relevance in a patient with multiple sclerosis showing its sensitivity to white matter lesions. In summary, DeepQSM can be used to determine the composition of myelin sheets of nerve fibers in the brain, and to assess quantitative information on iron homeostasis and its dysregulation, and will subsequently contribute to a better understanding of these biological processes in health and disease.

neuroscience↗

bioRxiv2017

Building a high-resolution in vivo minimum deformation average model of the human hippocampus

ObjectiveMinimum deformation averaging (MDA) procedures exploit the information contained in inter-individual variations to generate high-resolution, high-contrast models through iterative model building. However, MDA models built from different image contrasts reside in disparate spaces and their complementary information cannot be utilized easily. The aim of this work was to develop an algorithm for the non-linear alignment of two MDA models with different contrasts to create a high-resolution in vivo model of the human hippocampus with a spatial resolution of 300 m.\n\nMethodsA Turbo Spin Echo MDA model covering the hippocampus was contrast matched to a whole-brain MP2RAGE MDA model and aligned using cross-correlation and non-linear transformation. The contrast matching algorithm followed a global voxel location-based approach to estimate the relationship between intensity values of the two models. The performance of the algorithm was evaluated by comparing it to a non-linear registration obtained using mutual information without contrast matching. The complimentary information from both contrasts was then utilized in an automated hippocampal subfield segmentation pipeline.\n\nResultsThe contrast of the Turbo Spin Echo MDA model could successfully be matched to the MP2RAGE MDA model. Registration using cross correlation provided more accurate alignment of the models compared to a mutual information based approach. The segmentation using ASHS resulted in hippocampal subfield delineations that resembled the tissue boundaries observed in the Turbo Spin Echo MDA model.\n\nConclusionThe developed contrast matching algorithm facilitated the creation of a high-resolution multi-modal in vivo MDA model of the human hippocampus. This model can be used to improve algorithms for hippocampal subfield segmentation and could potentially support the early detection of neurodegenerative diseases.

neuroscience↗