Using machine learning-based lesion behavior mapping to identify anatomical networks of cognitive dysfunction: Spatial neglect and attention

Abstract

Previous lesion behavior studies primarily used univariate lesion behavior mapping techniques to map the anatomical basis of spatial neglect after right brain damage. These studies led to inconsistent results and lively controversies. Given these inconsistencies, the idea of a wide-spread network that might underlie spatial orientation and neglect has been pushed forward. In such case, univariate lesion behavior mapping methods might have been inherently limited in detecting the presumed network due to limited statistical power. By comparing various univariate analyses with multivariate lesion-mapping based on support vector regression, we aimed to validate the network hypothesis directly in a large sample of 203 newly recruited right brain damaged patients. If the exact same correction factors and parameter combinations (FDR correction and dTLVC for lesion size control) were used, both univariate as well as multivariate approaches uncovered the same complex network pattern underlying spatial neglect. At the cortical level, lesion location dominantly affected the temporal cortex and its borders into inferior parietal and occipital cortices. Beyond, frontal and subcortical gray matter regions as well as white matter tracts connecting these regions were affected. Our findings underline the importance of a right network in spatial exploration and attention and specifically in the emergence of the core symptoms of spatial neglect.

Keywords: Multivariate; Spatial attention; Stroke; Support vector regression; VLSM; Voxel-based lesion behavior mapping.

Figure 1:. Topography of brain lesions

A: Simple lesion overlap topography of all 203 patients. B: Lesion overlap topography showing only voxels within the voxel mask for statistical testing with at least 10 patients having a lesion. The colorbar indicates the number of overlapping lesions (the peak of N = 75 represents 37% of the total sample). Numbers above the slices indicate z-coordinates in MNI space.

Figure 2:. Estimation of best hyper-parameters C and γ

SVR-LSM parameter estimation results. Prediction Accuracy (A) and Reproducibility Index (B) (see Rasmussen et al., 2012; Zhang et al., 2014) are plotted for the different sets of C and γ parameters to find the optimal combination.

Figure 3:. Results of the multivariate lesion-behavior mapping

Support vector regression based multivariate lesion-symptom mapping results using data of 203 patients. Lesion volume correction was performed by applying dTLVC. A: Permutation-thresholded statistical map of SVR-LSM on CoC scores (FDR-corrected at q = 0.05, corresponding to a threshold of p < 0.0063), illustrating the anatomical regions significantly associated with the core deficit of spatial neglect. Significant clusters were interpreted according to the AAL atlas (Tzourio-Mazoyer et al., 2002) for grey matter regions and to the Juelich probabilistic cytoarchitectonic fiber tract atlas (Bürgel et al., 2006) as well as the tractography-based probabilistic fiber atlas by Thiebaut De Schotten et al. (2011) for white matter structures. B and C: three-dimensional renderings of the same map using the 3D-interpolation algorithm provided by MRIcron (http://people.cas.sc.edu/rorden/mricron/index.html; 8mm search depth) with sagittal view for B and inside view for C. Results of A, B and C are shown as 1-p. D: Thresholded β-parameter map showing only significant areas according to A. Abbreviations: SLF – superior longitudinal fasciculus; AF – arcuate fasciculus; ILF – inferior longitudinal fasciculus; IOF – inferior occipitofrontal fasciculus; SOF – superior occipitofrontal fasciculus; UF – Uncinate fasciculus.

Figure 4:. Results of the univariate lesion-behavior mapping using dTLVC

Mass-univariate lesion-symptom mapping results using data of 203 patients. Z-score maps are plotted for VLBM analysis with lesion volume correction by using dTLVC and FDR thresholded at q = 0.05, corresponding to a threshold of z > 3.3828. Significant clusters were interpreted according to the AAL atlas (Tzourio-Mazoyer et al., 2002) for grey matter regions and to the Juelich probabilistic cytoarchitectonic fiber tract atlas (Bürgel et al., 2006) as well as the tractography-based probabilistic fiber atlas by Thiebaut De Schotten et al. (2011) for white matter structures. B and C: three-dimensional renderings of the same map using the 3D-interpolation algorithm provided by MRIcron (http://people.cas.sc.edu/rorden/mricron/index.html; 8mm search depth) with sagittal view for B and inside view for C. Results of A, B and C are shown as 1-p. D: Thresholded β-parameter map showing only significant areas according to A. Abbreviations: SLF – superior longitudinal fasciculus; AF – arcuate fasciculus; ILF – inferior longitudinal fasciculus; IOF – inferior occipitofrontal fasciculus; SOF – superior occipitofrontal fasciculus; UF – Uncinate fasciculus.

Figure 5:. Results of all further univariate lesion-behavior mapping analyses

Mass-univariate lesion-symptom mapping results using data of 203 patients. Z-score maps are plotted for FWE permutation-thresholded as well as FDR-thresholded VLBM analyses with and without lesion volume correction on CoC scores. A: FWE permutation thresholded VLBM analysis without correction for lesion size at p < 0.05, corresponding to a threshold of z > 5.3475; B: FWE permutation thresholded VLBM analysis with correction for lesion size – by regressing lesion size out of behavior – at p < 0.05, corresponding to a threshold of z > 5.2251; C: FDR thresholded VLBM analysis without correction for lesion size at q = 0.05, corresponding to a threshold of z > 2.8607; D: FDR thresholded VLBM analysis with correction for lesion size – by regressing lesion size out of behavior – at q = 0.05, corresponding to a threshold of z > 4.4772.