Abstract:Objectives Sensitivity of noncontrast computed tomography (NCCT) in detecting hyperacute (<8 hours) and acute (<24 hours) cerebral infarction is low. We propose an automatic method to detect and localize ischemic infarct and to assess its volume from a single NCCT scan. Materials and Methods The method automatically determines attenuation value ranges of cerebrospinal fluid and white and gray matter, separates the brain scan into the left and right hemispheres, and by analyzing hemisphere attenuation value distributions using percentile difference ratios, it detects, localizes, and quantifies the infarct without its segmentation. The method performance was evaluated on 576 patients with clinically confirmed stroke through NCCT scans acquired at 4 centers to measure how it matched with that of experts in detection, localization, and assessment of infarct volume. The time from the onset of symptoms ranged from 1.5 to 72 hours for 450 scans and more than 72 hours for 82 scans, most with pathologic findings in addition to cerebral infarction; the time was unavailable for 44 scans. In addition, the method was compared with the novice’s (with 52 scans) and experienced readers’ infarct detection (with 21 × 2 scans) in early ischemia detection (with the time from the onset of symptoms ranging from 1.5 to 7 hours). Results The method matches 100% the expert’s infarct detection when chronic infarcts, leukoaraiosis cases, and infarct volumes less than 2 cm3 (determined by detection accuracy simulation) are excluded from the analysis. For all cases excluding infarct volumes less than 2 cm3, the method detection accuracy is 95.7%. Overall, the method detection accuracy is 83.2%. The early method detection accuracy (≤3 hours) is 78.4%. The novice detection accuracy is 27.8% (≤3 hours), 37.5% (3 < to ≤ 8 hours), and 77.8% (>8 hours), whereas the expert detection accuracy for these cases is 100%. Moreover, the method detected all 21 early infarcts, of which 15 were missed by the stroke experts and 14 of 15 were missed by a general radiologist. The method performs automatic analysis in approximately 7 seconds. Conclusions The results demonstrate potential benefits of our method for enhancing expert’s performance because it quickly localizes the infarct and detects cases missed by experts, and it is to be considered as an aid in the emergency department because it substantially outperforms novice readers (100% vs 27%) in infarct detection on NCCT.

Early Identification of Ischemic Stroke in Noncontrast Computed Tomography.

Identification Of Invisible Ischemic Stroke In Noncontrast Ct Based On Novel Two-Stage Convolutional Neural Network Model

Identification of early invisible acute ischemic stroke in non-contrast computed tomography using two-stage deep-learning model

Automatic identification of early ischemic lesions on non-contrast CT with deep learning approach

Localization of early infarction on non-contrast CT images in acute ischemic stroke with deep learning approach

Segmentation of Infarct Lesions and Prognosis Prediction for Acute Ischemic Stroke Using Non-Contrast CT Scans

Deep learning-based identification of acute ischemic core and deficit from non-contrast CT and CTA

Rapid Assessment of Acute Ischemic Stroke by Computed Tomography Using Deep Convolutional Neural Networks

White Matter High Signals Interfere with Noncontrast Computed Tomography in the Early Identification of Cerebral Infarction.

SFCN: Symmetric Feature Comparison Network for Detecting Ischemic Stroke Lesions on CT Images.

Target-based deep learning network surveillance of non-contrast computed tomography for small infarct core of acute ischemic stroke

An Improved Detection Algorithm for Ischemic Stroke NCCT Based on YOLOv5.

Computer-aided detection scheme for identification of hypoattenuation of acute stroke in unenhanced CT

Automated Cerebral Infarct Detection on Computed Tomography Images Based on Deep Learning

Automatic Detection, Localization, and Volume Estimation of Ischemic Infarcts in Noncontrast Computed Tomographic Scans

CNN Based Segmentation of Infarcted Regions in Acute Cerebral Stroke Patients From Computed Tomography Perfusion Imaging

Detecting the Early Infarct Core on Non-Contrast CT Images with a Deep Learning Residual Network

Fully Automatic Identification of Post-Treatment Infarct Lesions after Endovascular Therapy Based on Non-Contrast Computed Tomography

Predicting the clinical prognosis of acute ischemic stroke using machine learning: an application of radiomic biomarkers on non-contrast CT after intravascular interventional treatment

Automated Segmentation of Ischemic Stroke Lesions in Non-Contrast Computed Tomography Images for Enhanced Treatment and Prognosis

Value of 4D CT Angiography Combined with Whole Brain CT Perfusion Imaging Feature Analysis under Deep Learning in Imaging Examination of Acute Ischemic Stroke