ANALYSIS OF MULTIPLE SCLEROSIS LESIONS VIA SPATIALLY VARYING COEFFICIENTS

Bayesian Lesion Estimation with a Structured Spike-and-Slab Prior

Neural demyelination and brain damage accumulated in white matter appear as hyperintense areas on T2-weighted MRI scans in the form of lesions. Modeling binary images at the population level, where each voxel represents the existence of a lesion, plays an important role in understanding aging and inflammatory diseases. We propose a scalable hierarchical Bayesian spatial model, called BLESS, capable of handling binary responses by placing continuous spike-and-slab mixture priors on spatially-varying parameters and enforcing spatial dependency on the parameter dictating the amount of sparsity within the probability of inclusion. The use of mean-field variational inference with dynamic posterior exploration, which is an annealing-like strategy that improves optimization, allows our method to scale to large sample sizes. Our method also accounts for underestimation of posterior variance due to variational inference by providing an approximate posterior sampling approach based on Bayesian bootstrap ideas and spike-and-slab priors with random shrinkage targets. Besides accurate uncertainty quantification, this approach is capable of producing novel cluster size based imaging statistics, such as credible intervals of cluster size, and measures of reliability of cluster occurrence. Lastly, we validate our results via simulation studies and an application to the UK Biobank, a large-scale lesion mapping study with a sample size of 40,000 subjects.

GPU Accelerated Estimation of a Shared Random Effect Joint Model for Dynamic Prediction

In longitudinal cohort studies, it is often of interest to predict the risk of a terminal clinical event using longitudinal predictor data among subjects at risk by the time of the prediction. The at-risk population changes over time; so does the association between predictors and the outcome, as well as the accumulating longitudinal predictor history. The dynamic nature of this prediction problem has received increasing interest in the literature, but computation often poses a challenge. The widely used joint model of longitudinal and survival data often comes with intensive computation and excessive model fitting time, due to numerical optimization and the analytically intractable high-dimensional integral in the likelihood function. This problem is exacerbated when the model is fit to a large dataset or the model involves multiple longitudinal predictors with nonlinear trajectories. This challenge can be addressed from an algorithmic perspective, by a novel two-stage estimation procedure, and from a computing perspective, by Graphics Processing Unit (GPU) programming. The latter is implemented through PyTorch, an emerging deep learning framework. The numerical studies demonstrate that the proposed algorithm and software can substantially speed up the estimation of the joint model, particularly with large datasets. The numerical studies also concluded that accounting for nonlinearity in longitudinal predictor trajectories can improve the prediction accuracy in comparison to joint modeling that ignore nonlinearity.

Voxel-wise and spatial modelling of binary lesion masks: Comparison of methods with a realistic simulation framework

OBJECTIVES

White matter lesions are a very common finding on MRI in older adults and their presence increases the risk of stroke and dementia. Accurate and computationally efficient modelling methods are necessary to map the association of lesion incidence with risk factors, such as hypertension. However, there is no consensus in the brain mapping literature whether a voxel-wise modelling approach is better for binary lesion data than a more computationally intensive spatial modelling approach that accounts for voxel dependence.

METHODS

We review three regression approaches for modelling binary lesion masks including mass-univariate probit regression modelling with either maximum likelihood estimates, or mean bias-reduced estimates, and spatial Bayesian modelling, where the regression coefficients have a conditional autoregressive model prior to account for local spatial dependence. We design a novel simulation framework of artificial lesion maps to compare the three alternative lesion mapping methods. The age effect on lesion probability estimated from a reference data set (13,680 individuals from the UK Biobank) is used to simulate a realistic voxel-wise distribution of lesions across age. To mimic the real features of lesion masks, we propose matching brain lesion summaries (total lesion volume, average lesion size and lesion count) across the reference data set and the simulated data sets. Thus, we allow for a fair comparison between the modelling approaches, under a realistic simulation setting.

RESULTS

Our findings suggest that bias-reduced estimates for voxel-wise binary-response generalized linear models (GLMs) overcome the drawbacks of infinite and biased maximum likelihood estimates and scale well for large data sets because voxel-wise estimation can be performed in parallel across voxels. Contrary to the assumption of spatial dependence being key in lesion mapping, our results show that voxel-wise bias-reduction and spatial modelling result in largely similar estimates.

CONCLUSIONS

Bias-reduced estimates for voxel-wise GLMs are not only accurate but also computationally efficient, which will become increasingly important as more biobank-scale neuroimaging data sets become available.

Lesion probability mapping in MS patients using a regression network on MR fingerprinting

BACKGROUND

To develop a regression neural network for the reconstruction of lesion probability maps on Magnetic Resonance Fingerprinting using echo-planar imaging (MRF-EPI) in addition to [Formula: see text], [Formula: see text], NAWM, and GM- probability maps.

METHODS

We performed MRF-EPI measurements in 42 patients with multiple sclerosis and 6 healthy volunteers along two sites. A U-net was trained to reconstruct the denoised and distortion corrected [Formula: see text] and [Formula: see text] maps, and to additionally generate NAWM-, GM-, and WM lesion probability maps.

RESULTS

WM lesions were predicted with a dice coefficient of [Formula: see text] and a lesion detection rate of [Formula: see text] for a threshold of 33%. The network jointly enabled accurate [Formula: see text] and [Formula: see text] times with relative deviations of 5.2% and 5.1% and average dice coefficients of [Formula: see text] and [Formula: see text] for NAWM and GM after binarizing with a threshold of 80%.

CONCLUSION

DL is a promising tool for the prediction of lesion probability maps in a fraction of time. These might be of clinical interest for the WM lesion analysis in MS patients.

Tuber Locations Associated with Infantile Spasms Map to a Common Brain Network

OBJECTIVE

Approximately 50% of patients with tuberous sclerosis complex develop infantile spasms, a sudden onset epilepsy syndrome associated with poor neurological outcomes. An increased burden of tubers confers an elevated risk of infantile spasms, but it remains unknown whether some tuber locations confer higher risk than others. Here, we test whether tuber location and connectivity are associated with infantile spasms.

METHODS

We segmented tubers from 123 children with (n = 74) and without (n = 49) infantile spasms from a prospective observational cohort. We used voxelwise lesion symptom mapping to test for an association between spasms and tuber location. We then used lesion network mapping to test for an association between spasms and connectivity with tuber locations. Finally, we tested the discriminability of identified associations with logistic regression and cross-validation as well as statistical mediation.

RESULTS

Tuber locations associated with infantile spasms were heterogenous, and no single location was significantly associated with spasms. However, >95% of tuber locations associated with spasms were functionally connected to the globi pallidi and cerebellar vermis. These connections were specific compared to tubers in patients without spasms. Logistic regression found that globus pallidus connectivity was a stronger predictor of spasms (odds ratio [OR] = 1.96, 95% confidence interval [CI] = 1.10-3.50, p = 0.02) than tuber burden (OR = 1.65, 95% CI = 0.90-3.04, p = 0.11), with a mean receiver operating characteristic area under the curve of 0.73 (±0.1) during repeated cross-validation.

INTERPRETATION

Connectivity between tuber locations and the bilateral globi pallidi is associated with infantile spasms. Our findings lend insight into spasm pathophysiology and may identify patients at risk. ANN NEUROL 2021;89:726-739.

Mapping migraine to a common brain network

Inconsistent findings from migraine neuroimaging studies have limited attempts to localize migraine symptomatology. Novel brain network mapping techniques offer a new approach for linking neuroimaging findings to a common neuroanatomical substrate and localizing therapeutic targets. In this study, we attempted to determine whether neuroanatomically heterogeneous neuroimaging findings of migraine localize to a common brain network. We used meta-analytic coordinates of decreased grey matter volume in migraineurs as seed regions to generate resting state functional connectivity network maps from a normative connectome (n = 1000). Network maps were overlapped to identify common regions of connectivity across all coordinates. Specificity of our findings was evaluated using a whole-brain Bayesian spatial generalized linear mixed model and a region of interest analysis with comparison groups of chronic pain and a neurologic control (Alzheimer's disease). We found that all migraine coordinates (11/11, 100%) were negatively connected (t ≥ ±7, P < 10-6 family-wise error corrected for multiple comparisons) to a single location in left extrastriate visual cortex overlying dorsal V3 and V3A subregions. More than 90% of coordinates (10/11) were also positively connected with bilateral insula and negatively connected with the hypothalamus. Bayesian spatial generalized linear mixed model whole-brain analysis identified left V3/V3A as the area with the most specific connectivity to migraine coordinates compared to control coordinates (voxel-wise probability of ≥90%). Post hoc region of interest analyses further supported the specificity of this finding (ANOVA P = 0.02; pairwise t-tests P = 0.03 and P = 0.003, respectively). In conclusion, using coordinate-based network mapping, we show that regions of grey matter volume loss in migraineurs localize to a common brain network defined by connectivity to visual cortex V3/V3A, a region previously implicated in mechanisms of cortical spreading depression in migraine. Our findings help unify migraine neuroimaging literature and offer a migraine-specific target for neuromodulatory treatment.

MRI-based prediction of conversion from clinically isolated syndrome to clinically definite multiple sclerosis using SVM and lesion geometry

Neuroanatomical pattern classification using support vector machines (SVMs) has shown promising results in classifying Multiple Sclerosis (MS) patients based on individual structural magnetic resonance images (MRI). To determine whether pattern classification using SVMs facilitates predicting conversion to clinically definite multiple sclerosis (CDMS) from clinically isolated syndrome (CIS). We used baseline MRI data from 364 patients with CIS, randomised to interferon beta-1b or placebo. Non-linear SVMs and 10-fold cross-validation were applied to predict converters/non-converters (175/189) at two years follow-up based on clinical and demographic data, lesion-specific quantitative geometric features and grey-matter-to-whole-brain volume ratios. We applied linear SVM analysis and leave-one-out cross-validation to subgroups of converters (n = 25) and non-converters (n = 44) based on cortical grey matter segmentations. Highest prediction accuracies of 70.4% (p = 8e-5) were reached with a combination of lesion-specific geometric (image-based) and demographic/clinical features. Cortical grey matter was informative for the placebo group (acc.: 64.6%, p = 0.002) but not for the interferon group. Classification based on demographic/clinical covariates only resulted in an accuracy of 56% (p = 0.05). Overall, lesion geometry was more informative in the interferon group, EDSS and sex were more important for the placebo cohort. Alongside standard demographic and clinical measures, both lesion geometry and grey matter based information can aid prediction of conversion to CDMS.

Identifying therapeutic targets from spontaneous beneficial brain lesions

Brain damage can occasionally result in paradoxical functional benefit, which could help identify therapeutic targets for neuromodulation. However, these beneficial lesions are rare and lesions in multiple different brain locations can improve the same symptom. Using a technique called lesion network mapping, we show that heterogeneous lesion locations resulting in tremor relief are all connected to common nodes in the cerebellum and thalamus, the latter of which is a proven deep brain stimulation target for tremor. These results suggest that lesion network mapping can identify the common substrate underlying therapeutic lesions and effective therapeutic targets. Ann Neurol 2018;83:153-157.

Modelling the distribution of white matter hyperintensities due to ageing on MRI images using Bayesian inference

White matter hyperintensities (WMH), also known as white matter lesions, are localised white matter areas that appear hyperintense on MRI scans. WMH commonly occur in the ageing population, and are often associated with several factors such as cognitive disorders, cardiovascular risk factors, cerebrovascular and neurodegenerative diseases. Despite the fact that some links between lesion location and parametric factors such as age have already been established, the relationship between voxel-wise spatial distribution of lesions and these factors is not yet well understood. Hence, it would be of clinical importance to model the distribution of lesions at the population-level and quantitatively analyse the effect of various factors on the lesion distribution model. In this work we compare various methods, including our proposed method, to generate voxel-wise distributions of WMH within a population with respect to various factors. Our proposed Bayesian spline method models the spatio-temporal distribution of WMH with respect to a parametric factor of interest, in this case age, within a population. Our probabilistic model takes as input the lesion segmentation binary maps of subjects belonging to various age groups and provides a population-level parametric lesion probability map as output. We used a spline representation to ensure a degree of smoothness in space and the dimension associated with the parameter, and formulated our model using a Bayesian framework. We tested our algorithm output on simulated data and compared our results with those obtained using various existing methods with different levels of algorithmic and computational complexity. We then compared the better performing methods on a real dataset, consisting of 1000 subjects of the UK Biobank, divided in two groups based on hypertension diagnosis. Finally, we applied our method on a clinical dataset of patients with vascular disease. On simulated dataset, the results from our algorithm showed a mean square error (MSE) value of 7.27×10-5, which was lower than the MSE value reported in the literature, with the advantage of being robust and computationally efficient. In the UK Biobank data, we found that the lesion probabilities are higher for the hypertension group compared to the non-hypertension group and further verified this finding using a statistical t-test. Finally, when applying our method on patients with vascular disease, we observed that the overall probability of lesions is significantly higher in later age groups, which is in line with the current literature.

Spatially Adaptive Varying Correlation Analysis for Multimodal Neuroimaging Data

In this paper, we study a central problem in multimodal neuroimaging analysis, i.e., identification of significantly correlated brain regions between multiple imaging modalities. We propose a spatially varying correlation model and the associated inference procedure, which improves substantially over the common alternative solutions of voxel-wise and region-wise analysis. Compared with voxel-wise analysis, our method aggregates voxels with similar correlations into regions, takes into account spatial continuity of correlations at nearby voxels, and enjoys a much higher detection power. Compared with region-wise analysis, our method does not rely on any pre-specified brain region map, but instead finds homogenous correlation regions adaptively given the data. We applied our method to a multimodal positron emission tomography study, and found brain regions with significant correlation between tau and glucose metabolism that voxel-wise or region-wise analysis failed to identify. Our findings conform and lend additional support to prior hypotheses about how the two pathological proteins of Alzheimer's disease, tau and amyloid, interact with glucose metabolism in the aging human brain.

Voxelwise distribution of acute ischemic stroke lesions in patients with newly diagnosed atrial fibrillation: Trigger of arrhythmia or only target of embolism?

Objective

Atrial fibrillation (AF) is frequently detected after ischemic stroke for the first time, and brain regions involved in autonomic control have been suspected to trigger AF. We examined whether specific brain regions are associated with newly detected AF after ischemic stroke.

Methods

Patients with acute cerebral infarctions on diffusion-weighted magnetic resonance imaging were included in this lesion mapping study. Lesions were mapped and modeled voxelwise using Bayesian Spatial Generalised Linear Mixed Modeling to determine differences in infarct locations between stroke patients with new AF, without AF and with AF already known before the stroke.

Results

582 patients were included (median age 68 years; 63.2% male). AF was present in 109/582 patients [(18.7%); new AF: 39/109 (35.8%), known AF: 70/109 (64.2%)]. AF patients had larger infarct volumes than patients without AF (mean: 29.7 ± 45.8 ml vs. 15.2 ± 35.1 ml; p<0.001). Lesions in AF patients accumulated in the right central middle cerebral artery territory. Increasing stroke size predicted progressive cortical but not pontine and thalamic involvement. Patients with new AF had more frequently lesions in the right insula compared to patients without AF when stroke size was not accounted for, but no specific brain region was more frequently involved after adjustment for infarct volume. Controlled for stroke size, left parietal involvement was less likely for patients with new AF than for those without AF or with known AF.

Conclusions

In the search for brain areas potentially triggering cardiac arrhythmias infarct size should be accounted for. After controlling for infarct size, there is currently no evidence that ischemic stroke lesions of specific brain areas are associated with new AF compared to patients without AF. This challenges the neurogenic hypothesis of AF according to which a relevant proportion of new AF is triggered by ischemic brain lesions of particular locations.

Spatial modeling of multiple sclerosis for disease subtype prediction

Magnetic resonance imaging (MRI) has become an essential tool in the diagnosis and managing of Multiple Sclerosis (MS). Currently, the assessment of MS is based on a combination of clinical scores and subjective rating of lesion images by clinicians. In this work we present an objective 5-way classification of MS disease subtype as well as a comparison between three different approaches. First we propose two spatially informed models, a Bayesian Spatial Generalized Linear Mixed Model (BSGLMM) and a Log Gaussian Cox Process (LGCP). The BSGLMM accounts for the binary nature of lesion maps and the spatial dependence between neighboring voxels, and the LGCP accounts for the random spatial variation in lesion location. Both improve upon mass univariate analyses that ignore spatial dependence and rely on some level of arbitrarily defined smoothing of the data. As a comparison, we consider a machine learning approach based on multi-class support vector machine (SVM). For the SVM classification scheme, unlike previous work, we use a large number of quantitative features derived from three MRI sequences in addition to traditional demographic and clinical measures. We show that the spatial models outperform standard approaches with average prediction accuracies of up to 85%.