Prediction of Stroke Risk in Patients with Atrial Fibrillation Based on Chest CT Images

This study aims to develop and evaluate a deep learning framework that can automatically extract imaging features of the left atrial appendage in patients with atrial fibrillation and combine them with clinical features to predict the risk of ischemic stroke in these patients. The study intends to retrospectively collect clinical data (including patients' general information, medical history, laboratory tests, etc.) and chest CT images, as well as pulmonary vein CTV images (if available), from patients diagnosed with non-valvular atrial fibrillation between January 2018 and June 2024. The patients will be divided into stroke and non-stroke groups based on whether they have experienced an ischemic stroke, and a data analysis repository will be established. The dataset will be split into training and validation sets. Missing data will be handled, and data labeling will be performed on the pulmonary vein CTV sequence images and virtual non-contrast (VNC) sequence images. The left atrial morphology will be delineated, and a deep learning-based image segmentation network model will be developed to extract and select radiomic features for the prediction system.

Inclusion criteria

Diagnosed with atrial fibrillation by ECG, 24-hour Holter monitor, or recordable ECG monitor; atrial fibrillation confirmed by an implanted pacemaker or defibrillator, lasting at least 30 seconds Available chest CT images and complete clinical data.

Exclusion criteria

Incomplete clinical data or diagnosis of valvular atrial fibrillation (e.g., rheumatic heart valve disease, post-valve replacement) Poor-quality CT images that prevent complete assessment of left atrial appendage morphology Patients who have undergone left atrial appendage closure Patients who have had radiofrequency ablation or cardioversion with no evidence of recurrence post-procedure