HST Coverage
Heart disease has been the No. 2 cause of death in Taiwan, particularly acute myocardial infarction. Data and statistics of the National Health Insurance Administration show that more than 48000 people sought medical attention because of the condition throughout 2018 and every one out of eight patients was less than 50 years old. There is a tendency for the condition to affect also younger people.
It is somehow difficult to interpret myocardial infarction because it is an extremely critical condition that shall be determined and treated immediately. Typical symptoms of myocardial infarction include chest pain, shortness of breath, and cold sweat but some patients do not have them and hence are likely missed out in the beginning without even having the ECG done. According to prior data and statistics, before AI was introduced, for a patient that arrived at the hospital in the morning, it was unable to be detected despite the first ECG that took four minutes and the second one that occurred 23 minutes later. It was not until six hours that the blocked blood vessel was taken care of.
The emergency room (ER) of a hospital sees around 20 patients with myocardial infarction a month on average. These patients are discussed as a whole on a monthly basis because blood vessels that are blocked in the case of acute myocardial infarction need to be cleared as soon as possible. Any delay can lead to muscular impairment and give rise to more issues.
Useful AI Interprets by the Minute and Second 5000~6000 ECG Graphs a Month in ER
Generally speaking, it is without an issue if ECG graphs are interpreted by an experienced cardiologist; a cardiologist, however, is not available around the clock in the ER. As far as the ER is concerned, there are around 5 to 6 thousand ECG graphs produced a month and they are usually interpreted by the first-line ER doctors or residents. In other words, by training AI to be as specialized as a cardiologist, AI will be able to do the interpretation at any time. This is like having a specialist in the hospital around the clock. It also serves as a safety network that includes all typical and atypical patients to improve the safety and efficacy of medical care and its quality.
The design of the AI ECG interpretation procedure for automatic diagnosis of myocardial infarction takes into consideration the atypical symptoms of some patients with ST elevation myocardial infarction (STEMI) that result in a delay in having the ECG done or the failure to interpret the ECG graph correctly that leads to a delay in initiating cardiac catheterization. In the research and development of the AI ECG interpretation for automatic diagnosis of myocardial infarction, the cross-disciplinary team of the China Medical University Hospital (CMUH) started with interpretation of an ECG graph for arrhythmia and then acute myocardial infarction. It could render certain accuracy. The results were quite well. Meanwhile, the newly developed ASAP score and the AI-assisted STEMI interpretation system can remind ER doctors of having ECG done in a timely manner for this group of patients without typical symptoms of myocardial infarction even if they may have it and applying AI to quickly interpret each ECG graph prior to issuing a warning message in addition to the improved information system and a medical team of cardiologists and ER staff to ensure the most timely, sufficient, and effective medical care for patients in the ER.
Medicine-Industry-Academia Collaboration Speeds up R&D
The whole research team consists of members from the China Medical University Hospital (CMUH), the China Medical University (CMU), and Ever Fortune. Only clinicians know what are the issues. Therefore, the CMUH and the CMU are in charge of the first half of the development process where they find out how AI shall intervene. Ever Fortune, on the other hand, takes care of the second-half AI software research and development, clinical qualification, licensing, and marketing, etc. Multiple institutions in Taiwan have collaborated with the National Institutes of Health (NIH) in clinical trials.
The AI ECG interpretation system for automatic diagnosis of myocardial infarction is being investigated in a multi-center clinical trial in Taiwan in 2021. The collaboration also involves professors at the NIH in the US. It is ongoing at the moment.
For the 12 types of arrhythmia, the research team created a quick and accurate classification system (the US FDA permit is mostly about arrhythmia, too); it is released in an international journal (Can J Cardiol, 2020). Once the experience was acquired, it served as the basis for the development of an AI ECG interpretation system for STEMI (there are no such products approved by the FDA and the TFDA yet; AI-assisted interpretation of myocardial infarction is easier than arrhythmia and the standard changes make it easier for AI to learn); the system has an accuracy equivalent to that of a diagnosis rendered by a cardiologist.
Highly Precise 12-Lead ECG in 8 Minutes Effectively Reduces Physician Workload
The interpretation of ECG graphs for myocardial infarction is generally based on the variation of the so-called “ST-T segment” but under certain circumstances, interpretation may be difficult, such as early repolarization and artificial interference, etc. Generally speaking, it is advised in the acute STEMI treatment guidelines that from the time a patient steps into the ER to restoration of cardiac blood flows through catheterization, that is, the door-to-balloon time (D2B), it shall be within 90 minutes. The D2B time for many patients, nevertheless, exceeds 90 minutes mainly because of:
1. The atypical symptoms in some patients with STEMI, leading to a delay in getting the ECG done.
2. The failure to timely and accurately interpret the ECG graph to delay initiation of cardiac catheterization.
.png)
Arrhythmia is detected through single-lead ECG while myocardial infarction requires 12-lead ECG. The challenges are different. The AI Center, once spent lots of efforts in finding the most suitable model for the said training, too. The Apple Watch, for example, applies the single-lead and can detect atrial fibrillation. It has been approved by the FDA and TFDA, with an accuracy of about 80%. The other 20% is the false positive rate. It can monitor personal health. Nevertheless, a physician’s judgment and confirmation are required eventually.
Since AI was introduced to help with the ECG diagnosis system, it takes only eight minutes to complete a 12-lead ECG exam. The implementation outcome from the first four months already supported the quick diagnosis of STEMI. In the ER, a total of 21035 ECG procedures were performed and the system issued 213 text messages on AI-diagnosed STEMI; among them, 171 were confirmed with STEMI and 42 were false positive. The accuracy was up to 81%. It greatly reduced the workload of physicians.
.png)
AI Quickly and Precisely Interprets ECG Graphs for Automatic Diagnosis of Myocardial Infarction to Help Improve Patient Safety
For the AI model design, it is mainly about the adjustment and balance between “sensitivity” and “specificity”. When it is too sensitive, it increases false positivity and makes clinicians nervous at all times. Extreme specificity, on the other hand, is likely to lead to false negativity and hence a serious mis-diagnosis. Myocardial infarction is a critical condition that does not allow any false negativity and allows false positivity only to a certain extent. For the time being, the false positive rate is less than 20% and the accuracy is above 80%.
ECG graphs of acute myocardial infarction need to be interpreted “quickly” and “precisely”. Upon arrival of a patient, if he/she presents with the symptom of chest pain, ECG is performed directly. In cases of atypical symptoms, on the other hand, Top 5 atypical symptoms and high-risk factors (gender and age) are sorted out through the historical database of the chest pain center. Together with the “triage system”, those with higher risk scores are checked and the window showing “it is advised to perform ECG first within ten minutes” will automatically pop up.
According to the AI-assisted ECG interpretation system that has been developed by the Cardiovascular Center, with the 12-lead ECG, AI will help interpret the graph. If STEMI is determined, the ER doctor, cardiologist on duty, technologist, and PCI physician will be alerted simultaneously on their mobile phone and the ER will broadcast: “Acute STEMI＂. Then, the cardiologist will confirm again; if STEMI is consistently rendered, cardiac catheterization will be initiated. This solution further combines the ASAP score and the AI-assisted ECG interpretation to include patients with typical chest pain and atypical symptoms in the complete myocardial infarction diagnosis safety network, too.
ASAP Scoring System Effectively Shortens D2B Time
The ASAP scoring system that may be summarized through the hospital’s internal database by the age, (men >50; women >60), sex, and atypical presentation, including: 1. Altered consciousness; 2. Systemic malaise, weak limbs; 3. (Upper) Abdominal pain, vomiting. And past history, including: 1. hypertension, 2. diabetes, 3. known coronary artery disease, screens high-risk patients and applies AI to help with ECG-based diagnosis and to remind first-line ER doctors of performing ECG as soon as possible, interpreting the graph, and saving the lives of patients with myocardial infarction. When the ASAP score is ≧ 3, the window showing “it is advised to perform ECG first within ten minutes” will automatically pop up.
Following the introduction of the ASAP score, both the time needed for the ER to issue the ECG order and that for completion of ECG were reduced and the ratio of ECG procedures completed within ten minutes upon arrival at the ER climbed from 23.4% prior to introduction to 56.81%. Among the high-risk patients screened with the ASAP score, a total of three patients with STEMI had atypical presentation. Compared to the data between May 1 and May 25 prior to introduction, the duration of ECG was reduced from 29 to 5 minutes.
Following the introduction of AI and the ASAP score, the time from arrival at the ER to completion of ECG, that to contact with the cardiac catheterization unit, and the overall D2B time of patients with myocardial infarction were consistently shortened. Compared to the six months (a total of 120 patients) prior to introduction, the median D2B time of STEMI following introduction of the AI system was shortened from 61 to 53 minutes.
Analysis and comparison show that among all patients with myocardial infarction, it was completed within 90 minutes over 75% of patients. More than six months ago, it was about 88% on average and it climbed to 98% over the past four to five months. Another strength of the AI system is that when an ECG graph is not reviewed or can hardly be interpreted, AI will automatically give an alert; the time to initiation is obviously shortened so that all patients can complete it within 90 minutes. In the future, the system can be applied to other parts of Taiwan. Due to the drastic difference in the distribution of healthcare resources, there are only one or two cardiologists or even no cardiologists available in many outlying areas. Each month, there may only be ten days with doctors on duty. If ER doctors are incapable of interpreting the graphs, it is likely to result in a delay in detecting or treating the condition in a patient. The ASAP score and the AI-assisted STEMI ECG interpretation system helps ER doctors detect the condition and refer the patient early on.
Proper Internal ER Procedure Followed by External Promotion
At present, among the patients seeking medical attention for myocardial infarction in Taiwan, around 40% came in on an ambulance and 60% on their own. A proper medical procedure in the ER is the first step, followed by external promotion. 1. Simple ambulatory ECG equipment; 2. Personal portable ECG. Transmission to the cloud for AI diagnosis should be allowed in both cases. Once myocardial infarction is detected, the patient can begin cardiac catheterization directly upon approval at the hospital.
12-Lead is simple and can be done at home. Real-time feedback is possible with the combination of AI. For out-of-hospital application, it is taken into consideration resources available on an ambulance and people’s habit in seeking medical care so that non-professionals can use the device. With AI combined, real-time instructions in response can be provided in the first place to the ambulance and the individual as part of speedy management.
Through the smart Internet of Things (IoT), ECG graphs and patient locations are uploaded applying the AI AED and convenient ECG and the system will render the interpretation outcome confirmed by a cardiologist and the advised management. It helps greatly with the transfer of patients to a hospital or for the preparations in the cardiac catheterization unit of a hospital as it effectively reduces the time to treatment and the duration of myocardial ischemia and death rate of patients.
.png)