Cardiovascular disease is one of the major health concerns in Taiwan and
worldwide. The number of patients with cardiovascular disease has drama
tically increased over the years, with coronary artery disease (CAD) being t
he most significant condition. Traditional diagnostic methods for CAD are
expensive and resource-intensive. However, this technology utilizes artific
ial intelligence (AI)-enhanced electrocardiography (ECG) to improve the d
etection of CAD thereby enabling more accurate detection of asymptomat
ic coronary artery stenosis.This AI-based technology is a SaMD (software as medical device). Three p
atents are applied: (1) Method for selecting feature of ECG; (2) Method for
predicting blockage of coronary artery; (3) Method for diagnosing heart s
tate based on ECG. Moreover, we expand the output format to PDF vector
files, enhancing its commercial potential.
Regarding signal processing, it includes wavelet transform, statistical indic
ators, and personalized heartbeat intervals. Regarding feature engineerin
g, it includes various time intervals, slopes, and height differences betwee
n heartbeats to capture subtle abnormalities. Regarding model predictio
n, it utilizes a combination of deep learning and machine learning method
s, ensuring stability and accuracy.
In terms of product application, both XML and PDF files are applicable as i
nput. The interpretability of features aids clinicians in identifying abnorma
l locations on ECG. Finally, the performance was validated in two hospitals.
The results revealed that our AI technology has a superiorly better perfor
mance. Traditional resting ECG achieves to 50%-60% accuracy, and exercis
e ECG achieves to 70% accuracy. Remarkably, our AI algorithm achieves an
84%-90% accuracy of AUC.
Currently, CAD cannot be easily detected in high-risk asymptomatic patie
nts. Our AI algorithm can identify 80%-90% of asymptomatic patients wit
h normal ECG but actually having stenosis, thus enhancing the accuracy fo
r detecting