Groenewold, N. | bioRxiv Science

bioRxiv2018

Deep Learning for Quality Control of Subcortical Brain 3D Shape Models

We present several deep learning models for assessing the morphometric fidelity of deep grey matter region models extracted from brain MRI. We test three different convolutional neural net architectures (VGGNet, ResNet and Inception) over 2D maps of geometric features. Further, we present a novel geometry feature augmentation technique based on parametric spherical mapping. Finally, we present an approach for model decision visualization, allowing human raters to see the areas of subcortical shapes most likely to be deemed of failing quality by the machine. Our training data is comprised of 5200 subjects from the ENIGMA Schizophrenia MRI cohorts, and our test dataset contains 1500 subjects from the ENIGMA Major Depressive Disorder cohorts. Our final models reduce human rater time by 46-70%. ResNet outperforms VGGNet and Inception for all of our predictive tasks.

neuroscience↗

bioRxiv2018

Concordance Of Genetic Variation That Increases Risk For Tourette Syndrome And That Influences Its Underlying Neurocircuitry

BACKGROUNDThere have been considerable recent advances in understanding the genetic architecture of Tourette Syndrome (TS) as well as its underlying neurocircuitry. However, the mechanisms by which genetic variations that increase risk for TS - and its main symptom dimensions - influence relevant brain regions are poorly understood. Here we undertook a genome-wide investigation of the overlap between TS genetic risk and genetic influences on the volume of specific subcortical brain structures that have been implicated in TS.\n\nMETHODSWe obtained summary statistics for the most recent TS genome-wide association study (GWAS) from the TS Psychiatric Genomics Consortium Working Group (4,644 cases and 8,695 controls) and GWAS of subcortical volumes from the ENIGMA consortium (30,717 individuals). We also undertook analyses using GWAS summary statistics of key symptom factors in TS, namely social disinhibition and symmetry behaviour. SNP Effect Concordance Analysis (SECA) was used to examine genetic pleiotropy - the same SNP affecting two traits - and concordance - the agreement in SNP effect directions across these two traits. In addition, a conditional false discovery rate (FDR) analysis was performed, conditioning the TS risk variants on each of the seven subcortical and the intracranial brain volume GWAS. Linkage Disequilibrium Score Regression (LDSR) was used as validation of SECA.\n\nRESULTSSECA revealed significant pleiotropy between TS and putaminal (p=2x10-4) and caudal (p=4x10-4) volumes, independent of direction of effect, and significant concordance between TS and lower thalamic volume (p=1x10-3). LDSR lent additional support for the association between TS and thalamic volume (p=5.85x10-2). Furthermore, SECA revealed significant evidence of concordance between the social disinhibition symptom dimension and lower thalamic volume (p=1x10-3), as well as concordance between symmetry behaviour and greater putaminal volume (p=7x10-4). Conditional FDR analysis further revealed novel variants significantly associated with TS (p<8x10-7) when conditioning on intracranial (rs2708146, q=0.046; and rs72853320, q=0.035 and hippocampal (rs1922786, q=0.001 volumes respectively.\n\nCONCLUSIONThese data indicate concordance for genetic variations involved in disorder risk and subcortical brain volumes in TS. Further work with larger samples is needed to fully delineate the genetic architecture of these disorders and their underlying neurocircuitry.

genomics↗

bioRxiv2017

Machine Learning for Large-Scale Quality Control of 3D Shape Models in Neuroimaging

As very large studies of complex neuroimaging phenotypes become more common, human quality assessment of MRI-derived data remains one of the last major bottlenecks. Few attempts have so far been made to address this issue with machine learning. In this work, we optimize predictive models of quality for meshes representing deep brain structure shapes. We use standard vertex-wise and global shape features computed homologously across 19 cohorts and over 7500 human-rated subjects, training kernelized Support Vector Machine and Gradient Boosted Decision Trees classifiers to detect meshes of failing quality. Our models generalize across datasets and diseases, reducing human workload by 30-70%, or equivalently hundreds of human rater hours for datasets of comparable size, with recall rates approaching inter-rater reliability.

neuroscience↗