Computerized Detection of Early Ischemic Signs of Acute Stroke at Lentiform Nucleus in Unenhanced CT Using Deep Learning

Abstract

Recently, endovascular thrombectomy for acute ischemic stroke is gaining increasing attention. Identifying hypoattenuation of early ischemic changes on computerized tomography (CT) images is crucial for diagnosis. However, it is difficult to identify early ischemic changes with certainty. We present an atlas-based computerized method using a convolutional neural network (CNN) to identify early ischemic changes in the lentiform nucleus. The algorithm for this method consisted of anatomic standardization, setting of regions, creation of input images for classification, training on the CNN and classification of early ischemic changes. The method was applied to 50 patients with early ischemic change of acute stroke (<4.5 h) in the lentiform nucleus and 28 normal controls. As a result, we obtained a sensitivity of 90.0%, a specificity of 100％ and an accuracy of 93.6% for identifying early ischemic changes in the lentiform nucleus. These results indicate that this new method has the potential to accurately identify early ischemic changes in the lentiform nucleus in patients with acute ischemic stroke on CT images.