Distinguishing Stroke Patients with and Without Unilateral Spatial Neglect by Means of Clinical Features: a Tree-based Machine Learning Approach

Unilateral Spatial Neglect is a cognitive impairment of neuropsychological interest that is a consequence of stroke able to influence negatively the rehabilitation outcome of patients with stroke. The aim of the study is to explore the feasibility of machine learning to classify stroke patients with and without unilateral spatial neglect using clinical features. We performed the study using a machine learning approach by means the following tree-based algorithms: Decision Tree, Random Forest, Rotation Forest, AdaBoost of decision stumps and Gradient Boost tree using six clinical features both numerical and nominal: Montreal Cognitive Assessment, Functional Independence Measure scale, Barthel Index, aetiology, site of brain lesion and presence of hemiparesis at lower limbs. Tree-based Machine learning analysis achieved interesting results in terms of evaluation metrics scores; the best algorithm was Random Forest with an Accuracy, Sensitivity, Specificity, Precision and Area under the Receiver Operating Characteristic curve equal to 0.92, 0.83, 1.00, 1.00, 0.95 respectively. The study demonstrated the proposed combination of clinical features and algorithms represents a valuable approach to automatically classify stroke patients with and without Unilateral Spatial Neglect. The future investigations on enriched datasets will further confirm the potential application of this methodology as prognostic support to be chosen among those already implemented in the clinical field.