摘要
Introduction
Integrating imaging results across neuroimaging modalities and disorders is challenging. This difficulty is partly because imaging studies usually focus on a single disorder or modality, with only some with multimodal imaging. Studies involving three or more modalities is quite rare. Language being one of the most highly investigated cognitive function, language impairment has been studied almost across many neurological and psychiatric disorders irrespective of aetiology or modality, with previous research suggesting a shared neural substrate. This study reports a statistical framework to identify the shared neural substrate for language, integrating neuroimaging modalities (PET, DTI, VBM, fMRI, rs-fMRI, VLSM), reported statistics (Z, F, T, ALE) and aetiologies (vascular, neurodevelopmental, neurodegenerative, psychiatric) in select disorders (stroke, autism, Parkinson’s, schizophrenia, epilepsy).
Methods
Preprocessing included extraction of coordinates (in MNI152 space) from 49 studies (39 primary, 10 meta-analyses) across all five disorders, Z-score normalisation, peak cluster values modelled as 3D Gaussians scaled to the sample size. Data from 2,309 coordinates (epilepsy 967, stroke 650, schizophrenia 326, Parkinson’s 233, and autism 133) underwent Principal component analysis (PCA), followed by posterior probability maps using Variational Bayesian general linear models (VB-GLM) for each disorder and overall conjunction.
Results
PCA for all within-brain foci and effect strength (Z) was regressed on the first two components (PC1, PC2) separately for each condition. PC1 did not predict Z in any condition (overall p>0.09, R²=0.096; significant after Bonferroni correction) and weakly in stroke (β=0.003 Z mm-3, p=0.025, R²=0.011; not significant after correction). The joint model (PC1+PC2) did not improve on the single-component models (R²<0.097).
Conclusion
A convergent zone is identified for language processing centred around perisylvian regions for each neurological and psychiatric conditions for possible intervention and further investigations.
Keywords: Neurological and Neurodevelopmental disorders, Multimodal imaging, Variational Bayes, Principal Component Analysis, Perisylvian region
