Potential and limitations of computed tomography images as predictors of the outcome of ischemic stroke events: a review

Abstract

The prediction of functional outcome after a stroke remains a relevant, open problem. In this article, we present a systematic review of approaches that have been proposed to predict the most likely functional outcome of ischemic stroke patients, as measured by the modified Rankin scale. Different methods use a variety of clinical information and features extracted from brain computed tomography (CT) scans, usually obtained at the time of hospital admission. Most studies have concluded that CT data contains useful information, but the use of this information by models does not always translate into statistically significant improvements in the quality of the predictions.

Deep Learning-Based Extraction of Biomarkers for the Prediction of the Functional Outcome of Ischemic Stroke Patients

“…For a more in-depth analysis of these studies, as well as a more thorough review of the various previously published image-only and hybrid models studies, please refer to Oliveira et al [5].…”

“…which used CTAs and other patient variables as input. All these (image-only and hybrid) experiments were modeled as a binary classification problem by splitting the mRS target variable, 3 months after stroke into good outcome (mRS ≤ 2) and poor outcome (mRS > 2) classes, as is usually done in the literature [5].…”

“…The patient's functional outcome is commonly considered three months after onset and measured by the modified Rankin Scale (mRS), which is an integer scale that goes from zero to six, where zero corresponds to full independence and six corresponds to death [4,5]. Several models have been proposed by the medical and machine learning (ML) communities to predict this variable.…”

“…Several models have been proposed by the medical and machine learning (ML) communities to predict this variable. In this work, we categorized these studies into tabular, image-only and hybrid approaches, following the organization proposed by Oliveira et al [5]. As shown in Figure 1, each approach gets its name from the type of data ingested:…”