Functional Contributions Derived from the Game-theoretical Analysis of Brain Damage after Stroke
Melissa Zavaglia (Dept. Comput. Neurosci. Univ. Med. Ctr. Hamburg-Eppendorf; Jacobs Univ. Bremen), Nils D Forkert (Dept. Comput. Neurosci. Univ. Med. Ctr. Hamburg-Eppendorf), Bastian Cheng (Dept. Neurology, Univ. Med. Ctr. Hamburg-Eppendorf), Götz Thomalla (Dept. Neurology, Univ. Med. Ctr. Hamburg-Eppendorf), Claus C Hilgetag (Dept. Comput. Neurosci. Univ. Med. Ctr. Hamburg-Eppendorf; Jacobs Univ. Bremen; Boston Univ.)
MSA is a rigorous theoretical method to infer functional contributions from behavioral performance after multiple lesions, treating brain regions as players in a coalition game. For each coalition of regions, the system’s performance (inverse of NIHSS) is measured. MSA then derives each region’s contribution to behavioral function from analyzing all possible conﬁgurations of intact and lesioned regions.
Using a large multi-centre dataset of stroke patients , we investigated 9 bilateral regions of interest (ROI), deﬁned by the MNI152 atlas : caudate, cerebellum, insula, putamen, thalamus, frontal, occipital, parietal and temporal lobes. The overlap (in %) between infarct lesions registered to the atlas and each ROI was calculated for each patient. The resulting dataset was composed of 159 cases with diﬀerent patterns of lesioned ROIs and associated NIHSS. Lesion percentages were thresholded (eg, 0.16%) to obtain binary sets of lesioned and intact regions, and diﬀerent classiﬁers (shown here: regtree) were trained on the available set of lesion conﬁgurations to predict the performance of the remaining, unknown conﬁgurations, required for MSA.
Fig. 1 shows relative, unitless contribution values in a gray-scale map. Highest contributions are found in the left frontal lobe, left thalamus, followed by bilateral caudate, while other regions make weaker positive or negative contributions. The results are in line with the known pivotal role of these regions for basic brain function as measured by the NIHSS. Further contributions can be computed based on individual NIHSS components, to obtain a speciﬁc map for each task, such as language or attention, and provide detailed insights for rehabilitation.
Supported by ERA-NET NEURON (MZ, CCH)
 Brott 1989 Stroke
 Keinan 2004 Neur Comp
 Thomalla 2011 Lancet Neur
 Collins 1995 HBM