The value of artificial intelligence in the detection of early cerebral changes in acute stroke using non-contrast CT scans

Abstract

Background: Over the last few years, there has been increasing interest in the use of deep learning algorithms to assist with abnormality detection on medical images. Aim and Objective: was to assess the value of artificial intelligence in the detection of early cerebral changes in acute stroke using non-contrast CT scans. Patient & Methods: this cross sectional study included 1095 patients distributed across both training and validation, as well as a separate test set. Using 48 hours follow up non contrast CT images as the main reference standard to diagnose the acute ischemic stroke at the initial CT images & AI. Axial scanning extending from base of the skull up to the vertex with coronal & sagittal reformate images. Results: There were no statistically significant differences found among the diagnosis results of the first and second radiologist’s diagnosis and the AI system diagnosis (p > 0.05). Conclusion: In spite CAD system has established fair accuracy, the need of more accurate algorithm is necessary to determine if it can replicate non contrast CT and radiologist observations.