SleepMI: An AI-based screening algorithm for myocardial infarction using nocturnal electrocardiography.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Myocardial infarction (MI) necessitates early diagnosis for effective treatment and reduced mortality.
• Existing methods for MI detection vary, highlighting the need for novel, automated screening tools.
• Nocturnal electrocardiography (ECG) from polysomnography (PSG) offers a potential data source for MI detection during sleep.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI) algorithm, termed sleep-myocardial infarction (sleepMI), for automatic MI screening.
• To leverage nocturnal ECG findings from PSG data for robust MI detection.
• To assess the performance of the sleepMI algorithm in identifying MI events.

Main Methods:

• The sleepMI algorithm was developed using representation and ensemble learning, incorporating deep convolutional neural networks and LightGBM models.
• Nocturnal ECG signals were extracted from 2,691 participants (360 MI patients) in the Sleep Heart Health Study, segmented into 30-s intervals.
• The model was trained, validated, and tested on a large dataset comprising over 1.4 million ECG segments.

Main Results:

• The sleepMI algorithm achieved high performance metrics: 99.38% precision, 99.38% recall, and 99.38% F1-score.
• The overall mean accuracy for automatic MI screening using nocturnal single-lead ECG was 99.387%.
• The model demonstrated the capability to detect MI events using both conventional 12-lead ECG and polysomnographic ECG recordings.

Conclusions:

• The developed sleepMI algorithm provides a highly accurate and reliable method for automatic MI screening using nocturnal ECG data.
• This AI-driven approach holds significant potential for early MI detection, potentially improving patient outcomes and reducing mortality.
• The study validates the utility of integrating AI with PSG data for cardiovascular disease screening.