Can coronary artery blockages be predicted? The latest artificial intelligence technology has identified key features associated with coronary artery blockages. By utilizing ECG signal bands, it can predict blockages and prevent missed early treatment opportunities! Development using 561 features combined with logistic regression
Analysis Professor Chi-Hsiao Yeh of Linkou Chang Gung Memorial Hospital, in collaboration with the National Health Research Institutes and Acer Group, has developed the "Coronary Artery Blockage Intelligent Detection System." This technology enhances ECG analysis using AI, employing the xGBoost model to analyze 561 features from a 12-lead ECG. It combines these with logistic regression analysis to identify multiple key features related to coronary artery blockages, including R-wave, T-wave, and ST-segment changes. This significantly improves accuracy and reduces the need for invasive examinations.
The operational mechanism of this technology is as follows:
Subtle Changes in ECG Signals and Patterns: ECG signals reflect changes in the heart's electrical activity, and coronary artery blockages cause abnormalities in these signals. The AI model can capture these subtle changes to identify the presence of blockages.
Abnormal Calcium Cycles in Damaged Myocardium: Coronary artery blockages lead to insufficient blood supply to the myocardium, affecting calcium ion metabolism in heart cells. By analyzing ECG features, the AI model can infer abnormal changes in calcium ions in myocardial cells, indirectly indicating the extent of blockages. Achieving 90% AUC and detection sensitivity exceeding 80%
Professor Chi-Hsiao Yeh noted that compared to traditional ECG and exercise ECG accuracy rates, this AI detection system achieved an AUC of 84%-90% and a detection sensitivity of 79%-83% in tests conducted at two hospitals. Moreover, traditional clinical processes can only identify 13% of coronary artery stenosis patients, whereas this system can identify an additional 80% of these patients, significantly improving early detection rates.
This technology, classified as medical device software, is protected by multiple patents, including those for ECG signal processing and multi-stage machine learning algorithms. This ensures its advanced nature and uniqueness, marking a significant advancement in the field of ECG diagnostics. It holds the potential to enhance early detection and treatment of heart diseases, playing a crucial role in safeguarding public cardiovascular health.
Resource (mandarin)
