A Computational Tool for Enhancing Ischemic Stroke in Computed Tomography Examinations

Abstract

Stroke is a cardio-vascular disease that currently ranks in the fifth position among all causes of death worldwide. Computed tomography is the first radiologic examination performed in emergency decisions to diagnose stroke. The earliest signs of ischemic stroke are quite subtle in CT, thus image-processing tools can be used to enhance ischemic areas and to aid physicians during diagnosis. This study aimed to enhance the ischemic stroke visual perception in computed tomography examinations. A cohort of 45 exams were used during this study, with 28 patients previously diagnosed with ischemic stroke and 17 control patients. Stroke cases were obtained within 4.5 h of symptom onset and with mean NIHSS of 13.6 ± 5.5. The complete series of non-enhanced images were obtained in DICOM format and all processing was performed in Matlab software R2017a. The main steps of the computed algorithm were as follows: an image averaging was performed to reduce the noise and redundant information within each slice; then a variational decomposition model was applied to keep the relevant component for our analysis; then three different segmentation methods were used to enhance the ischemic stroke area. The segmentation methods used were expectation maximization method, K-means and mean-shift. We determined a test to evaluate the performance of six observers (physicians) in a clinical environment with and without the aid of enhanced images. According to the opinion of the observers who participated in this study the enhanced images were particularly useful when displayed together with the original images. The overall sensitivity of the observer’s analysis changed after the evaluation of the enhanced images with the expectation maximization method. The overall specificity also increased. The improvement was even more remarkable for the three least experienced physicians.