Stroke prediction algorithm based on 3D convolutional neural network for CT scans in patients with atrial fibrillation

Stroke, commonly associated with cerebral infarction, is a severe condition caused by disrupted blood flow to the brain. Atrial Fibrillation (AF), a prevalent cardiac arrhythmia, significantly increases the risk of stroke, with AF patients having a tenfold higher likelihood of stroke compared to the general population. Early detection of AF and the presence of blood clots is crucial for stroke prevention. In this study, we propose an AI-assisted diagnostic system based on a 3D convolutional neural network (CNN) trained on cardiac CT images to predict the risk of stroke in AF patients. Compared to traditional 2D CNN models, the proposed 3D CNN approach effectively captures 3D spatial features of cardiac structures, resulting in improved accuracy and performance. The 3D CNN model achieved an impressive accuracy of 92.92% and an AUC of 0.97 on the test set. The findings highlight the potential of AI-assisted diagnosis and the significance of utilizing cardiac CT images in enhancing cardiovascular disease diagnosis. This approach offers promising opportunities to improve accuracy, efficiency, and clinical decision-making in stroke prevention. Future research should focus on expanding the dataset, optimizing the model architecture, and integrating additional clinical data further to enhance the predictive performance of the AI model.