Deep Learning Empowered Spectral CT for Precision Diagnosis of Acute Ischemic Stroke (Preprint)

Abstract

BACKGROUND Acute Ischemic Stroke (AIS) diagnosis and management require timely and accurate assessment. Artificial Intelligence (AI) coupled with Spectral Computed Tomography (CT) imaging holds promise for enhancing diagnostic precision, thrombolysis efficacy evaluation, and long-term prognosis prediction. OBJECTIVE This study aims to evaluate the effectiveness of AI utilizing Deep Learning (DL) models in improving early AIS diagnosis, predicting thrombolysis outcomes, and providing robust prognostic tools for enhanced patient outcomes, aiding clinical decision-making. METHODS Clinical and brain imaging data from AIS patients were subjected to rigorous quality control measures. A Convolutional Neural Network (CNN), specifically a Residual Network (ResNet), was employed for image segmentation. The Densely Connected Convolutional Network (Dense-Net) model was trained using a cross-entropy loss function and the Adam optimizer, with model performance continuously monitored in real-time. RESULTS The DL model, leveraging Spectral CT imaging data, significantly increased the accuracy, sensitivity, and specificity of early AIS diagnosis. Notably, the model exhibited favorable prognostic performance, facilitating individualized treatment strategies correlated with clinical severity and long-term outcomes. CONCLUSIONS AI technology based on Spectral CT imaging presents a valuable tool for enhancing early AIS diagnosis, optimizing thrombolysis assessments, and accurately predicting long-term prognoses. This non-invasive, user-friendly approach has the potential for widespread adoption, marking a significant advancement in AI applications for AIS diagnosis and treatment.

Abstract

Related Papers