Atrial Fibrillation Detection Using MEMS Accelerometer Based Bedsensor

Abstract

Atrial fibrillation (AFib) is the most common cardiac arrhythmia, affecting eventually up to a quarter of the population. The purpose of this small scale clinical study was to validate the usability of MEMS accelerometer based bedsensor for detection of AFib. A Murata accelerometer based ballistocardiogram bedsensor was attached under the hospital bed magnetically and measurement data was recorded from 20 AFib patients and 15 healthy volunteers, mainly females. The recording time was up 30 minutes. The sensor built-in algorithms automatically extracted features such as heart rate (HR), heart rate variability (HRV), relative stroke volume (SVOL), signal strength (SS) and whether the patient is in bed or not. We calculated median values for each feature HR, HRV, SVOL and SS, and investigated whether it is possible to separate AFib from healthy with these features or their combinations. Areas under the curve (AUC) were 0.98 for full length signals and 0.85 for 3 min signal segments using random forest (RF) classifier corresponding to sensitivity and specificity of 100% and 93.3% for full length signals and 90% and 80% for 3 min signals. We conclude, that based on our pilot results, the Murata bedsensor is able to detect AFib, and seems to be a promising technology for long-term monitoring of AFib at home settings as it requires only one-time installation and operational time can be up to years and even tens of years.

Recent Research for Unobtrusive Atrial Fibrillation Detection Methods Based on Cardiac Dynamics Signals: A Survey

“…The obtained BCG signal corresponded to the upper part of the back of the patients’ body and was used to distinguish AF from non-atrial fibrillation (NAF) [ 20 , 21 ]. Koivisto et al also applied the Murata EMFi sensor with a sampling rate of 125 Hz to collect BCG signals and detected AF from SR [ 22 ]. Moreover, Panula et al collected BCG signals with a three-axis accelerometer and a three-axis gyroscope to detect AF [ 23 ].…”

“…For the segmentation, different frame lengths were selected, which could describe the rhythmical changes in AF periods. The frame length of BCG signal was usually set as 5 s [ 22 ], 24 s [ 20 , 21 ], 30 s [ 2 , 17 , 18 ] or 60 s [ 16 ]. The frame length of SCG signal was generally set as 5 s [ 22 , 23 ], 10 s [ 4 , 33 , 34 , 35 , 36 ] or 12.5 s [ 30 , 31 ], 30 s [ 41 , 45 ].…”

“…The frame length of BCG signal was usually set as 5 s [ 22 ], 24 s [ 20 , 21 ], 30 s [ 2 , 17 , 18 ] or 60 s [ 16 ]. The frame length of SCG signal was generally set as 5 s [ 22 , 23 ], 10 s [ 4 , 33 , 34 , 35 , 36 ] or 12.5 s [ 30 , 31 ], 30 s [ 41 , 45 ]. Moreover, the frame length of PPG signal was commonly set as 1.66 s [ 46 ], 5 s [ 32 ], 10 s [ 57 ], 15 s [ 50 ], 17 s [ 51 ], 17.1 s [ 52 ], 30 s [ 40 , 43 , 44 , 45 , 47 , 54 ], or 60 s [ 42 ].…”

“…For AF detection, Eerikäinen et al used RF with 30 s PPG segments and achieved a sensitivity of 93.6% and a specificity of 88.2% [ 43 ]. Koivisto et al used the RF classifier to classify 5 s BCG signals for detecting atrial fibrillation and compared it with the SVM classifier [ 22 ]. RF performed better than SVM, obtaining 100% sensitivity and 93.3% specificity.…”

A novel multi-axis pendulum sandwich structure accelerometer

“…Another direction of the development of MEMS accelerometers is high performance and high reliability, among which the overall performance of capacitive MEMS accelerometers is better [9]. Representative companies of high-performance accelerometers include Colibrys [10] and Murata [11]. Since the mechanical noise of the accelerometer determines its performance, and the most important factor affecting the mechanical noise is the size of the sensitive mass [12], both companies use the pendulum structure for acceleration detection.…”