Structural connectivity-based predictors of cognitive impairment in stroke patients attributable to aging.
Barbora Rehák BučkováDavid KalaJakub KořenekVeronika MatuškováVojtěch KumpoštLenka SvobodováJakub OtáhalAntonín ŠkochVlastimil ŠulcAnna OlšerováMartin VyhnálekPetr JanskýAleš TomekPetr MarusičPřemysl JiruškaJaroslav HlinkaPublished in: PloS one (2023)
Despite the rising global burden of stroke and its socio-economic implications, the neuroimaging predictors of subsequent cognitive impairment are still poorly understood. We address this issue by studying the relationship of white matter integrity assessed within ten days after stroke and patients' cognitive status one year after the attack. Using diffusion-weighted imaging, we apply the Tract-Based Spatial Statistics analysis and construct individual structural connectivity matrices by employing deterministic tractography. We further quantify the graph-theoretical properties of individual networks. The Tract-Based Spatial Statistic did identify lower fractional anisotropy as a predictor of cognitive status, although this effect was mostly attributable to the age-related white matter integrity decline. We further observed the effect of age propagating into other levels of analysis. Specifically, in the structural connectivity approach we identified pairs of regions significantly correlated with clinical scales, namely memory, attention, and visuospatial functions. However, none of them persisted after the age correction. Finally, the graph-theoretical measures appeared to be more robust towards the effect of age, but still were not sensitive enough to capture a relationship with clinical scales. In conclusion, the effect of age is a dominant confounder especially in older cohorts, and unless appropriately addressed, may falsely drive the results of the predictive modelling.
Keyphrases
- white matter
- cognitive impairment
- multiple sclerosis
- working memory
- diffusion weighted imaging
- resting state
- end stage renal disease
- functional connectivity
- chronic kidney disease
- newly diagnosed
- ejection fraction
- convolutional neural network
- contrast enhanced
- magnetic resonance imaging
- deep learning
- magnetic resonance
- brain injury
- blood brain barrier
- machine learning