Atrial Fibrillation Detection Based on Ballistocardiogram
Wen, Xin (2020-06-05)
Atrial Fibrillation Detection Based on Ballistocardiogram
Wen, Xin
(05.06.2020)
Julkaisu on tekijänoikeussäännösten alainen. Teosta voi lukea ja tulostaa henkilökohtaista käyttöä varten. Käyttö kaupallisiin tarkoituksiin on kielletty.
suljettu
Julkaisun pysyvä osoite on:
https://urn.fi/URN:NBN:fi-fe2020070246782
https://urn.fi/URN:NBN:fi-fe2020070246782
Tiivistelmä
Atrial fibrillation (AF) is one of the most common arrhythmia, and the prevalence increases with age. If AF is not treated in time, it will lead to the occurrence of many other cardiovascular diseases and strokes, which seriously affect people's quality of life and increase the medical burden of the society. With the trend of population aging in China intensifying, the diagnosis and treatment of AF is a public health issue worthy of attention. AF often occurs suddenly without warning. It is difficult to detect in the absence of a heart function detecting instrument, resulting in the patient not receiving treatment in time. Therefore, it is necessary to have a real-time monitoring device for heart function which is suitable for daily life, to find out the abnormality of the heart and prompt in time.
Ballistocardiogram (BCG) is a contactless method for heart activity detection. The signal is mainly generated due to changes in human gravity during blood circulation. The BCG signal can be easily accessed without attaching a sensor to the human body. This method of detection can obtain the activity of the heart in the case where the subject does not feel the measurement state, and the long-term use does not burden the subject. Thus, compared with other cardiovascular testing techniques, the BCG has the advantages of noninvasion, no direct contact and easy operation.
Based on the society and clinical needs, we used the BCG, a new cardiac function detection technology, to detect and discriminate the occurrence of AF in the topic. The main contents and innovation include:
(1) We proposed a BCG signal AF recognition algorithm based on machine learning. The algorithm minimizes the negative impact of sample variability on classification while highlights the difference between normal and AF BCG signals by transforming the pre-processed BCG signal into BCG energy signals. According to the characteristics of AF and non-AF signals, we extracted 16 features on the transformed signals in several aspects such as amplitude, peak interval and smoothness. Through these features we used several traditional machine learning algorithms to classify signals. We tested the algorithm with a total length of 2,915 minutes of BCG signal. The sensitivity, precision and accuracy of the best classifier were 0.968, 0.928 and 0.945, respectively.
(2) Design and implement a hardware acquisition system for BCG signals. According to the characteristics of the BCG signal, PVDF piezoelectric sensors are used for acquisition. The hardware system is capable of successfully acquiring the subject's BCG signal and performing preliminary filtering and amplification processing. The collected signals can be transmitted to the computer via USB.
(3) We proposed a new beat-to-beat heart rate detection algorithm and explored the feasibility of detecting BCG heart rate in patients with AF. The proposed algorithm performs heart rate detection based on the correlation of adjacent heartbeats. BCG signals with a total length of 145 minutes from 18 subjects (sinus rhythms) were used for the algorithm test. The precisions of heartbeat interval detection with an error less than 30 ms, 40 ms, and 50 ms were achieved to 86.92%, 95.22%, and 97.63%, respectively. However, this algorithm is not feasible in AF BCG signals.
Ballistocardiogram (BCG) is a contactless method for heart activity detection. The signal is mainly generated due to changes in human gravity during blood circulation. The BCG signal can be easily accessed without attaching a sensor to the human body. This method of detection can obtain the activity of the heart in the case where the subject does not feel the measurement state, and the long-term use does not burden the subject. Thus, compared with other cardiovascular testing techniques, the BCG has the advantages of noninvasion, no direct contact and easy operation.
Based on the society and clinical needs, we used the BCG, a new cardiac function detection technology, to detect and discriminate the occurrence of AF in the topic. The main contents and innovation include:
(1) We proposed a BCG signal AF recognition algorithm based on machine learning. The algorithm minimizes the negative impact of sample variability on classification while highlights the difference between normal and AF BCG signals by transforming the pre-processed BCG signal into BCG energy signals. According to the characteristics of AF and non-AF signals, we extracted 16 features on the transformed signals in several aspects such as amplitude, peak interval and smoothness. Through these features we used several traditional machine learning algorithms to classify signals. We tested the algorithm with a total length of 2,915 minutes of BCG signal. The sensitivity, precision and accuracy of the best classifier were 0.968, 0.928 and 0.945, respectively.
(2) Design and implement a hardware acquisition system for BCG signals. According to the characteristics of the BCG signal, PVDF piezoelectric sensors are used for acquisition. The hardware system is capable of successfully acquiring the subject's BCG signal and performing preliminary filtering and amplification processing. The collected signals can be transmitted to the computer via USB.
(3) We proposed a new beat-to-beat heart rate detection algorithm and explored the feasibility of detecting BCG heart rate in patients with AF. The proposed algorithm performs heart rate detection based on the correlation of adjacent heartbeats. BCG signals with a total length of 145 minutes from 18 subjects (sinus rhythms) were used for the algorithm test. The precisions of heartbeat interval detection with an error less than 30 ms, 40 ms, and 50 ms were achieved to 86.92%, 95.22%, and 97.63%, respectively. However, this algorithm is not feasible in AF BCG signals.