Search

Many key findings in neuroimaging studies involve similarities between brain maps, but statistical methods used to measure these findings have varied. Current state‐of‐the‐art methods involve comparing observed group‐level brain maps (after averaging intensities at each image location across multiple subjects) against spatial null models of these group‐level maps. However, these methods typically make strong and potentially unrealistic statistical assumptions, such as covariance stationarity. To address these issues, in this article we propose using subject‐level data and a classical permutation testing framework to test and assess similarities between brain maps. Our method is comparable to traditional permutation tests in that it involves randomly permuting subjects to generate a null distribution of intermodal correspondence statistics, which we compare to an observed statistic to estimate a p‐value. We apply and compare our method in simulated and real neuroimaging data from the Philadelphia Neurodevelopmental Cohort. We show that our method performs well for detecting relationships between modalities known to be strongly related (cortical thickness and sulcal depth), and it is conservative when an association would not be expected (cortical thickness and activation on the n‐back working memory task). Notably, our method is the most flexible and reliable for localizing intermodal relationships within subregions of the brain and allows for generalizable statistical inference., We propose using a classical permutation testing framework to study intermodal correspondence using subject‐level data while requiring minimal statistical assumptions. We compare our method to previous approaches involving spatial null modeling of group‐level brain maps and illustrate and discuss the flexibility of our method for localizing intermodal relationships within subregions of the brain.

A central challenge of medical imaging studies is to extract quantitative biomarkers that characterize pathology or predict disease outcomes. In high-resolution, high-quality magnetic resonance images (MRI), state-of-the-art approaches have performed well. However, such methods may not translate to low resolution, lower quality images acquired on MRI scanners with lower magnetic field strength. Therefore, in low-resource settings where low field scanners are more common and there is a shortage of available radiologists to manually interpret MRI scans, it is essential to develop automated methods that can accommodate lower quality images and augment or replace manual interpretation. Motivated by a project in which a cohort of children with cerebral malaria were imaged using 0.35 Tesla MRI to evaluate the degree of diffuse brain swelling, we introduce a fully automated framework to translate radiological diagnostic criteria into image-based biomarkers. We integrate multi-atlas label fusion, which leverages high-resolution images from another sample as prior spatial information, with parametric Gaussian hidden Markov models based on image intensities, to create a robust method for determining ventricular cerebrospinal fluid volume. We further propose normalized image intensity and texture measurements to determine the loss of gray-to-white matter tissue differentiation and sulcal effacement. These integrated biomarkers are found to have excellent classification performance for determining severe cerebral edema due to cerebral malaria.

Normal and pathologic neurobiological processes influence brain morphology in coordinated ways that give rise to structural covariance patterns across brain regions and individuals. We present a mega-analysis of structural covariance with magnetic resonance imaging of 50,699 healthy and diseased individuals (12 studies, 130 sites, and 12 countries) over their lifespan (ages 5 through 97). Patterns of structural covariance (PSC) were highly heritable (0.05< h2 -log10[p-value] > 8.8): 1245 previously unreported, and 69% of them independently replicated (-log10[p-value] = 4.5). Associations revealed an imaging phenotypic landscape between 2003 PSCs and 49 clinical and cognitive traits at multiple scales. We constructed machine learning-derived individualized imaging signatures for various disease diagnoses using PSC features at multiple scales, suggesting that disease effects on the brain were better manifested in a multi-scale continuum than on any single scale. Experimental results were integrated into the Multi-scale Structural Imaging Covariance (MuSIC) atlas and made publicly accessible through the BRIDGEPORT web portal (https://www.cbica.upenn.edu/bridgeport/). Our results reveal strong associations between brain structural covariance, genetics, and clinical phenotypes, supporting that PSCs can serve as an endophenotypic anatomic dictionary in future research.

Many key findings in neuroimaging studies involve similarities between brain maps. While several statistical procedures have been proposed to test correspondence between maps, there remains no consensus on the correct framing of a null hypothesis or a suitable testing approach. We propose a simple yet powerful permutation-based testing procedure for assessing similarities between two modalities using subject-level data. Our proposed method is similar to traditional permutation procedures in that it involves randomly permuting subjects to generate a null distribution. However, it differs from other recently proposed methods that have involved spherical rotations of the cortical surface or spatial autocorrelation-preserving “surrogate” maps of the brain, which depend on strong and potentially unrealistic statistical assumptions. To address these issues, we first demonstrate in simulated data that our method is conservative in terms of type I error and has high power. Next, we illustrate that our method performs well for assessing intermodal relationships from multimodal magnetic resonance imaging data from the Philadelphia Neurodevelopmental Cohort. The proposed test rejects the null hypothesis for modalities for which there is known interdependence in structure (cortical thickness and sulcal depth) but not in cases where an association would not be predicted biologically (cortical thickness and activation on the n-back working memory task). In contrast to previous methods, our approach does not depend on strong statistical assumptions other than the independence of subjects. Notably, our method is the most flexible for analyzing intermodal correspondence within subregions of the brain and has the greatest potential to be used for generalizable statistical inference.