Advantages
This device allows;
- diagnosis in significantly less time than a conventional Holter ECG (24 hours), reducing the burden on patients and physicians.
- testing of asymptomatic subjects and detection of potential patients through medical check-up.
Background and Technology
Atrial fibrillation (AF) is one of the most common causes of stroke or myocardial infarction, and early detection of AF is important. Conventional diagnosis of paroxysmal AF is made by continuously recording electrocardiogram (ECG) waveforms over a period of 24 hours to several days with a Holter-type ECG, and picking up ECG waveforms that appear to be during an attack. In some cases, a portable electrocardiograph is placed on the anterior chest when symptoms are present to record the ECG waveforms on the spot, and the waveforms are examined by a physician later. Both methods place a burden on the patient and the physician. Recently, it has become possible to detect arrhythmia using smartwatches with ECG functions, which can be used to assist in AF diagnosis, and efforts are being made to reduce the burden on patients and physicians. However, these efforts are insufficient. Furthermore, how to detect asymptomatic latent patients (half of patients with AF have no symptoms) is an important issue.
We developed an AF severity classification model based on heart rate variability (HRV) using XGBoost with ECG data from 75 AF-suspect subjects as training data. This model was able to diagnose AF severity with an accuracy of 86.2% correct, 1.53% false positive, and 11.4% false positive rate for high-level abnormalities. Therefore, if this model can be further developed and put into practical use, the burden on patients and physicians in AF diagnosis will be reduced, asymptomatic subjects can be tested easily, and potential patients can be detected by incorporating the model into medical examinations.
Reference and Patent
- The 36th Annual Conference of the Japanese Society for Artificial Intelligence, 2022
Development of severality diagnosis model of atrial fibrillation using XGBoost from electrocardiogram - Patent pending
Principal Investigator
Koichi FUJIWARA
(Tokai National Higher Education and Research System, Nagoya University Graduate School of Engineering, Laboratory of Human & Process Systems)
Current Stage and Next Step
- Currently in the process of developing machine learning models based on clinical data and demonstrating basic efficacy in diagnosis using clinical data.
- Plans to further increase clinical data and validate performance.
- Seeking partners to develop clinical programs or devices for commercialization.
Project No:BK-04289