XÂY DỰNG MÔ HÌNH TRÍ TUỆ NHÂN TẠO ĐỂ TẦM SOÁT RUNG NHĨ TRÊN DỮ LIỆU LỚN ĐIỆN TÂM ĐỒ LƯU ĐỘNG TẠI BỆNH VIỆN NGUYỄN TRÃI

Abstract

Background: Atrial fibrillation (AF) is a common clinical arrhythmia with significant implications. Current AF screening is performed using algorithm-based software, which has many limitations and burdens healthcare staff when analyzing big electrocardiography (ECG) data. Artificial intelligence (AI) is a promising tool to support AF screening. Objectives: To develop an AI model with sufficient capability to screen for atrial fibrillation and assess the performance of the AI model in screening for atrial fibrillation using real-world 24-hour Holter ECG datasets. Methods: A retrospective study of big 24-hour Holter ECG data was conducted at Nguyen Trai Hospital from March to September 2024. The AI model was built based on the ResNet architecture in deep learning. Results: From 1089 training Holter ECG datasets, 3218 AF segments (1785953 seconds) and 2631 non-AF segments (486775 seconds) were selected to build the AI model. On 400 evaluation datasets, the optimal AI model achieved a sensitivity of 100% and specificity of 80% in screening for AF. Conclusion: The AI model developed by Nguyen Trai Hospital demonstrates potential for practical application in effectively screening for AF in big ECG data.