The aim of this work was to quantitatively model cross-sectional relationships

The aim of this work was to quantitatively model cross-sectional relationships between structural connectome disruptions caused by cerebral infarction and measures of clinical performance. (ChaCo) score. ChaCo scores were utilized because they can be calculated using routinely collected clinical MRIs. Partial Least Squares Regression (PLSR) was used to predict various acute impairment and activity measures from ChaCo scores and patient demographics. Statistical methods of cross-validation bootstrapping and multiple comparisons correction were implemented to minimize over-fitting and Type I errors. Multiple linear regression models based on lesion volume and lateralization information were constructed for comparison. All models based on connectivity disruption had lower Merck SIP Agonist Akaike Information Criterion and almost all had better goodness-of-fit values (R2:0.26-0.92) than models based on lesion characteristics (R2:0.06-0.50). Confidence intervals of PLSR coefficients identified brain regions S100A4 important in predicting each clinical assessment. Appropriate mapping of eloquent functions i.e. language and motor and replication of results across pathologies provided validation of this method. Models of complex functions provided new insights into brain-behavior relationships. In addition to the potential applications in prognostication and rehabilitation development this quantitative approach provides insight into the structural networks underlying complex functions like activities of daily living and cognition. Quantitative analysis of big data will be invaluable in understanding complex brain-behavior relationships. techniques in lesion-mapping studies to enhance our understanding of eloquent cortical areas such as those responsible for language and motor functions [Butler et al. 2014 Hope et al. 2013 Phan et al. 2010 general intelligence[Barbey et al. 2012 Gl?scher et al. 2010 and neglect[Mort et al. 2003 However the anatomical substrates underlying performance in more general tasks like basic activities of daily living and more complex behaviors that arise from distributed brain networks are not as fully understood. Machine learning techniques applied to neuroscientific “big data” sets will be central to understanding these complex brain-behavior relationships. One such machine learning technique is the method of partial least squares regression (PLSR)[Wold 1982 PLSR has been applied in the field of neuroimaging in previous studies of brain-behavior relationships mostly in the analysis of practical MRI[Hay et al. 2002 Itier et al. 2004 (observe Krishnan et. al. 2011 for a review). For example one study[Phan et al. 2010 investigated the effect of infarct size and location on engine and language function at a voxel-wise level using a logistic version of PLSR. This work and that of others [Kuceyeski et al. 2011 Menezes et al. 2007 have reinforced the well-established notion that the location of tissue damage is a key factor determining the attendant practical deficit i.e. sign or symptom. Advanced neuroimaging techniques and quantitative methods e.g. voxel-based morphometry [Ashburner and Friston 2000 and voxel-based lesion-symptom mapping[Bates et al. 2003 can be used to map voxel-wise guidelines to behavior. However it isn’t just a lesion’s location in gray matter (GM) that is important since damage can also disrupt WM tracts that connect GM areas. This disruption of the brain’s structural contacts in turn affects function[Johansen-Berg et al. 2010 Puig et al. 2013 and possibly recovery[Crofts et al. 2011 vehicle Hees et al. 2014 Consequently Merck SIP Agonist we hypothesized that models based on steps of the brain’s structural connectome disruption due Merck SIP Agonist to a lesion’s size and location will result in more accurate predictions of medical assessments than a model based on lesion characteristics. To test this hypothesis we used the recently developed Network Changes (NeMo) Tool [Kuceyeski et al. 2013 in conjunction with PLSR to link patterns of disruption in the brain’s structural connectome to steps of various cognitive motor language and daily living activities inside a cohort of individuals with ischemic stroke much like [Kuceyeski et al. in press]. The NeMo Tool quantifies the amount of connectivity disruption that a given cortical Merck SIP Agonist or subcortical region has incurred due to a given WM lesion by using Merck SIP Agonist normal subjects’ structural connectivity information. This tool is attractive because it uses MRI sequences regularly. Merck SIP Agonist