Lesion severity and critical eloquent brain areas for ischemic stroke outcome prediction

Introduction Acute ischemic stroke is one of the leading causes of disability globally, requiring the best-integrated approach between prevention, intervention, and therapies aiming to avoid the worse outcome scenario. A careful analysis of potential clinical predictors after stroke under the current data science paradigms can significantly improve prognosis reliability and lead to the anticipation of the most suitable rehabilitation program to be adopted, with crucial benefits to avoid severe disabilities. Among the predictors used, the Alberta Stroke Program Early CT Score (ASPECTS) deserves special attention, given its standardized assessment by evaluating computed tomography (CT) scans. Despite its widespread use, the ASPECTS has some limitations since it does not consider different eloquent brain regions impact on stroke patients’ functional outcome or even the lesion severity of the respective area. Objective This study aims to identify the eloquent brain areas most related to the patients’ prognosis through machine learning techniques, considering a binary (affected or non-affected region — as ranked under ASPECTS paradigm) and a weighted (3-level damage severity degree) approach. The stroke outcome prediction performance was investigated and compared with results obtained under the classical ASPECTS. Methods We applied a wrapper feature selection based on sequential forward strategy combined with a linear discriminant analysis classifier considering: (1) the binary lesion approach for investigating the role of critical eloquent brain areas; (2) the weighted lesion approach for investigating the role of lesion severity; (3) the ASPECTS predictor performance for benchmark comparison. Results We identified six promising areas and observed that the weighted approach had the best performance among the binary option and ASPECTS scenarios. Conclusion Despite being more complex, critical eloquent brain areas and damage severity can play a crucial role in improving clinical prediction in ischemic stroke prognosis, which deserves careful attention in the context of the current data science techniques.