Machine learning as an aid to predicting clinical outcome after stroke

Abstract

Numerous models have been developed to predict mortality in spontaneous intracerebral hemorrhage (ICH), which is one of the types of stroke with high mortality [1] [2]. Prediction of the clinical outcome in ICH is a significant help to the neurologist in making decisions about the optimal treatment of the patient and personalized therapy. In this paper, neuro-fuzzy models for predicting mortality after spontaneous ICH based on initial clinical parameters have been developed and compared with published models based on artificial neural networks and logistic regression. A set of data on patients with spontaneous ICH published in a study [3] has been used, where patients were treated for a five-year period at a university clinical center belonging to tertiary health care. Patients older than 18 years of age who had evidence of spontaneous ICH on computed tomography of the brain have been considered. Data on 411 patients (199 men and 212 women), with mean age of 67.35 years, have been analyzed, of which 256 (62.29%) patients passed away in hospital during treatment and 155 (37.71%) patients survived. The developed neuro-fuzzy models have shown superiority compared to standard logistic regression models, while the accuracy of classification has been worse compared to the model based on artificial neural networks published in [3]. On the other hand, the developed neuro-fuzzy models have other advantages that have been discussed in the paper.