Differentiating Between Brain Hemorrhage and Contrast

Sun Yat-Sen Memorial Hospital of Sun Yat-Sen University500 enrolled

Overview

The goal of this observational study is to use artificial intelligence to differentiate cerebral hemorrhage from contrast agent extravasation after mechanical revascularization in ischemic stroke. The main question it aims to answer is: Whether artificial intelligence can help differentiate brain hemorrhage from contrast agent extravasation. Patients with intracranial high-density lesions on CT scans within 24h after mechanical revascularization will be included. Expected to enroll 500 patients. The type of high-density lesion is determined according to dual-energy CT images or follow-up images. Patients will be divided into training group, validation and testing groups by stratified random sampling (6:2:2). After the images and the image labels are obtained, deep learning artificial intelligence will be used to learn the image characteristics and establish a diagnostic model, and the model performance and generalization ability will be evaluated.

Study Type

OBSERVATIONAL

Enrollment

500

Conditions

Ischemic Stroke

Eligibility

Sex: ALLMin age: 18 YearsMax age: 100 Years

Medical Language ↔ Plain English

Inclusion Criteria: (1) patients underwent non-enhanced head CT after mechanical vascularization; (2) initial post-operative non-enhanced head CT was performed within 24 h after mechanical vascularization; and (3) intracranial hyper-intensity, which was defined as an objectively higher density than the surrounding grey or white matter in the parenchyma or higher density than cerebrospinal fluid in ventricles and cisterns, could be seen on the initial non-enhanced head CT after mechanical vascularization. Exclusion Criteria: (1) the follow-up time of non-enhanced head CT after mechanical vascularization was less than 24 h; (2) artifacts (e.g. metal artifacts or motion artifacts) affected the hyper-intensity in CT images; and (3) patients underwent craniotomy after mechanical vascularization, which made it difficult to identify the area of hyper-intensity.

Outcomes

Primary Outcomes

Develop a deep learning model to differentiate brain hemorrhage from contrast agent extravasation, and evaluate the model performance and generalization ability

The accuracy, sensitivity, specificity, precision, and recall of the model will be calculated, and confusion matrix will be display.

Time frame: 2024-12