Lesion to symptom mapping (LSM) is a crucial tool for understanding the causality of brain-behavior relationships. The analyses are typically performed by applying statistical methods on individual brain voxels (VLSM), a method called the mass-univariate approach. Several authors have shown that VLSM suffers from limitations that may decrease the accuracy and reliability of the findings, and have proposed the use of multivariate methods to overcome these limitations. In this study, we propose a multivariate optimization technique known as sparse canonical correlation analysis for neuroimaging (SCCAN) for lesion to symptom mapping. To validate the method and compare it with mass-univariate results, we used data from 131 patients with chronic stroke lesions in the territory of the middle cerebral artery, and created synthetic behavioral scores based on the lesion load of 93 brain regions (putative functional units). LSM analyses were performed with univariate VLSM or SCCAN, and the accuracy of the two methods was compared in terms of both overlap and displacement from the simulated functional areas. Overall, SCCAN produced more accurate results - higher dice overlap and smaller average displacement - compared to VLSM. This advantage persisted at different sample sizes (N = 20-131) and different multiple comparison corrections (false discovery rate, FDR; Bonferroni; permutation-based family wise error rate, FWER). These findings were replicated with a fully automated SCCAN routine that relied on cross-validated predictive accuracy to find the optimal sparseness value. Simulations of one, two, and three brain regions showed a systematic advantage of SCCAN over VLSM; under no circumstance could VLSM exceed the accuracy obtained with SCCAN. When considering functional units composed of multiple brain areas VLSM identified fewer areas than SCCAN. The investigation of real scores of aphasia severity (aphasia quotient and picture naming) showed that SCCAN could accurately identify known language-critical areas, while VLSM either produced diffuse maps (FDR correction) or few scattered voxels (FWER correction). Overall, this study shows that a multivariate method, such as, SCCAN, outperforms VLSM in a number of scenarios, including functional dependency on single or multiple areas, different sample sizes, different multi-area combinations, and different thresholding mechanisms (FWER, Bonferroni, FDR). These results support previous claims that multivariate methods are in general more accurate than mass-univariate approaches, and should be preferred over traditional VLSM approaches. All the methods described in this study are available in the newly developed LESYMAP package for R.

A thalamic network associated with seizure recurrence may already be established presurgically. Thalamic hubness can serve as a potential biomarker of surgical outcome, outperforming the clinical characteristics commonly used in epilepsy surgery centers.

The gold standard for identifying stroke lesions is manual tracing, a method that is known to be observer dependent and time consuming, thus impractical for big data studies. We propose LINDA (Lesion Identification with Neighborhood Data Analysis), an automated segmentation algorithm capable of learning the relationship between existing manual segmentations and a single T1-weighted MRI. A dataset of 60 left hemispheric chronic stroke patients is used to build the method and test it with k-fold and leave-one-out procedures. With respect to manual tracings, predicted lesion maps showed a mean dice overlap of 0.696±0.16, Hausdorff distance of 17.9±9.8mm, and average displacement of 2.54±1.38mm. The manual and predicted lesion volumes correlated at r=0.961. An additional dataset of 45 patients was utilized to test LINDA with independent data, achieving high accuracy rates and confirming its cross-institutional applicability. To investigate the cost of moving from manual tracings to automated segmentation, we performed comparative lesion-to-symptom mapping (LSM) on five behavioral scores. Predicted and manual lesions produced similar neuro-cognitive maps, albeit with some discussed discrepancies. Of note, region-wise LSM was more robust to the prediction error than voxel-wise LSM. Our results show that, while several limitations exist, our current results compete with or exceed the state-of-the-art, producing consistent predictions, very low failure rates, and transferable knowledge between labs. This work also establishes a new viewpoint on evaluating automated methods not only with segmentation accuracy but also with brain-behavior relationships. LINDA is made available online with trained models from over 100 patients.

Voxel-based lesion-symptom mapping (VLSM) is an important method for basic and translational human neuroscience research. VLSM leverages modern neuroimaging analysis techniques to build on the classic approach of examining the relationship between location of brain damage and cognitive deficits. Testing an association between deficit severity and lesion status in each voxel involves very many individual tests and requires statistical correction for multiple comparisons. Several strategies have been adapted from analysis of functional neuroimaging data, though VLSM faces a more difficult trade-off between avoiding false positives and statistical power (missing true effects). Non-parametric, permutation-based methods are generally preferable because they do not make assumptions that are likely to be violated by skewed distributions of behavioral deficit (symptom) scores and by the necessary spatial contiguity of stroke lesions. We used simulated and real deficit scores from a sample of approximately 100 individuals with left hemisphere stroke to evaluate two such permutation-based approaches. Using permutation to set a minimum cluster size identified a region that systematically extended well beyond the true region, even under the most conservative settings tested here, making it ill-suited to identifying brainbehavior relationships. In contrast, generalizing the standard permutation-based family-wise error correction approach provided a principled way to balance false positives and false negatives. An implementation of this continuous permutation-based FWER correction method is available at 1 Address correspondence to Daniel Mirman, dan@danmirman.org. Portions of this work were previously described in arXiv:1606.00475 [stat.AP].

The severity of post-stroke aphasia and the potential for recovery are highly variable and difficult to predict. Evidence suggests that optimal estimation of aphasia severity requires the integration of multiple neuroimaging modalities and the adoption of new methods that can detect multivariate brain-behavior relationships. We created and tested a multimodal framework that relies on three information sources (lesion maps, structural connectivity, and functional connectivity) to create an array of unimodal predictions which are then fed into a final model that creates “stacked multimodal predictions” (STAMP). Cross-validated predictions of four aphasia scores (picture naming, sentence repetition, sentence comprehension, and overall aphasia severity) were obtained from 53 left hemispheric chronic stroke patients (age: 57.1±12.3yrs, post-stroke interval: 20 months, 25 female). Results showed accurate predictions for all four aphasia scores (correlation true vs. predicted: r=0.79–0.88). The accuracy was slightly smaller but yet significant (r=0.66) in a full split cross-validation with each patient considered as new. Critically, multimodal predictions produced more accurate results that any single modality alone. Topological maps of the brain regions involved in the prediction were recovered and compared with traditional voxel-based lesion-to-symptom maps, revealing high spatial congruency. These results suggest that neuroimaging modalities carry complementary information potentially useful for the prediction of aphasia scores. More broadly, this study shows that the translation of neuroimaging findings into clinically useful tools calls for a shift in perspective from unimodal to multimodal neuroimaging, from univariate to multivariate methods, from linear to non-linear models, and, conceptually, from inferential to predictive brain mapping.

No abstract

In temporal lobe epilepsy (TLE), determining the hemispheric specialization for language before surgery is critical to preserving a patient's cognitive abilities post-surgery. To date, the major techniques utilized are limited by the capacity of patients to efficiently realize the task. We determined whether resting-state functional connectivity (rsFC) is a reliable predictor of language hemispheric dominance in right and left TLE patients, relative to controls. We chose three subregions of the inferior frontal cortex (pars orbitalis, pars triangularis and pars opercularis) as the seed regions. All participants performed both a verb generation task and a resting-state fMRI procedure. Based on the language task, we computed a laterality index (LI) for the resulting network. This revealed that 96% of the participants were left-hemisphere dominant, although there remained a large degree of variability in the strength of left lateralization. We tested whether LI correlated with rsFC values emerging from each seed. We revealed a set of regions that was specific to each group. Unique correlations involving the epileptic mesial temporal lobe were revealed for the right and left TLE patients, but not for the controls. Importantly, for both TLE groups, the rsFC emerging from a contralateral seed was the most predictive of LI. Overall, our data depict the broad patterns of rsFC that support strong versus weak left hemisphere language laterality. This project provides the first evidence that rsFC data may potentially be used on its own to verify the strength of hemispheric dominance for language in impaired or pathologic populations.

Pre-surgical evaluation of patients with temporal lobe epilepsy (TLE) relies on information obtained from multiple neuroimaging modalities. The relationship between modalities and their combined power in predicting the seizure focus is currently unknown. We investigated asymmetries from three different modalities, PET (glucose metabolism), MRI (cortical thickness), and diffusion tensor imaging (DTI; white matter anisotropy) in 28 left and 30 right TLE patients (LTLE and RTLE). Stepwise logistic regression models were built from each modality separately and from all three combined, while bootstrapped methods and split-sample validation verified the robustness of predictions. Among all multimodal asymmetries, three PET asymmetries formed the best predictive model (100% success in full sample, >95% success in split-sample validation). The combinations of PET with other modalities did not perform better than PET alone. Probabilistic classifications were obtained for new clinical cases, which showed correct lateralization for 7/7 new TLE patients (100%) and for 4/5 operated patients with discordant or non-informative PET reports (80%). Metabolism showed closer relationship with white matter in LTLE and closer relationship with gray matter in RTLE. Our data suggest that metabolism is a powerful modality that can predict seizure laterality with high accuracy, and offers high value for automated predictive models. The side of epileptogenic focus can affect the relationship of metabolism with brain structure. The data and tools necessary to obtain classifications for new TLE patients are made publicly available.