Deep learning identifies brain structures that predict cognition and explain heterogeneity in cognitive aging

The volume and the distribution of premorbid white matter hyperintensities: Impact on post‐stroke aphasia

“…The contribution of the frontal white matter in healthy and pathological ageing has been a topic of extensive research (Brickman et al, 2012 ; Fjell et al, 2017 ; Schneider et al, 2022 ). Recent analyses of microstructural indices in large cohort samples confirmed that frontal white matter is singularly vulnerable to microstructural changes as we age and that these changes are predictive of worse cognitive performance and further cognitive decline (Poulakis et al, 2021 ; Saboo et al, 2022 ; Vemuri et al, 2021 ). From a network perspective, segments of the frontal white matter may constitute ‘key hubs’ (Stam, 2014 ) or ‘bottlenecks’ (Griffis et al, 2017 ), that is, brain structures that are preferentially afflicted across disorders, with damage to them being disproportional associated with psycho‐neurological disturbance (van den Heuvel & Sporns, 2019 ).…”

“…topic of extensive research (Brickman et al, 2012;Fjell et al, 2017;Schneider et al, 2022). Recent analyses of microstructural indices in large cohort samples confirmed that frontal white matter is singularly vulnerable to microstructural changes as we age and that these changes are predictive of worse cognitive performance and further cognitive decline (Poulakis et al, 2021;Saboo et al, 2022;Vemuri et al, 2021). From a network perspective, segments of the frontal white matter may constitute 'key hubs' (Stam, 2014) or 'bottlenecks' (Griffis et al, 2017), that is, brain structures that are preferentially afflicted across disorders, with damage to them being disproportional Coupled with the absence of an identified relationship between WMH lesion load and other CC segments, this finding suggests that frontal callosal connections may play a strategic role in cognitive networks, with minor disruptions resulting in behavioural consequences.…”

Cerebral thickness reveals sex differences in verbal and visuospatial memory

“…They are based on the presupposition that there is a linear relationship between the independent and dependent variables. However, linear models cannot account for complex interactions between brain structures and cognitions (Nelson et al, 2009; Saboo et al, 2022). Although nonlinear relationships between brain and cognition can be explored by extending the linear model, such as using the generalized linear model, this approach also still requires making assumptions about the form of nonlinear relationships prior to analysis.…”

“…However, there are some problems with using these machine learning methods in cognitive neuroscience research. Among them, the accuracy and interpretability of the models are the two most important concerns for researchers (Linardatos et al, 2020; Murdoch et al, 2019; Saboo et al, 2022). An accurate model with good interpretability and without assuming the pattern of relationship between variables in advance, can help deepen our understanding of the more fundamental relationship between brain and behavior.…”

“…To this end, the present study first conducted a validation exercise to investigate the differences in two types of memory performance (verbal and visuospatial memory) and cortical thickness between females and males. Although cortical thickness and cortical area are heritable and have both distinct genetic determinants and different developmental trajectories (Ge et al, 2021), cortical thickness is more closely related to intelligence and cognitive function, and reflects the cytoarchitectural characteristics (Saboo et al, 2022). Then, a gradient boosting technique-extreme gradient boosting (XGBoost) (T. Chen & Guestrin, 2016), which was able to produce explainable and interpretable predicitons, was used to develop predictive models between cortical thickness and memory performance in male and female populations, respectively.…”