Automatic Ballistocardiogram (BCG) Beat Detection Using a Template Matching Approach Adviser: Ji-Jer Huang Presenter: Zhe-Lin Cai Date:2014/12/24 30th Annual International IEEE EMBS Conference Vancouver, British Columbia, Canada, August 20-24, 2008 J. H. Shin, B. H. Choi, Y. G. Lim, D. U. Jeong and K. S. Park

Outline Introduction Method Result Discussion 2

Introduction Ballistocardiography (BCG) ― Cardiac and respiratory evaluation ― Non-invasive method ― Strength of myocardial contraction ― Condition of the heart 3 Photo Source :

Introduction Ventricle contraction ―ECG Q-R-S complex ―BCG I-J-K complex ECG, BCG principles ―ECG records the heart in nerve conduction arising from potential changes graphics ―BCG defined as a method by which body vibrations caused by heart activity are recorded 4 Photo Source :

Introduction BCG: ―Advantages: Non-contact Non-conscious Security ―Disadvantages: Motion artifact signal 5

Introduction This paper suggests a beat detection method for ballistocardiogram (BCG) from an unconstrained cardiac signal monitoring devices ―The goal of the method is extraction of J peak without ECG synchronization 6 Photo Source :

Introduction The analyzed systems were Chee et al (2005) method using balancing tube and air-mattress for unconstrained measurement system loadcell type BCG measurement system and EMFi-film measurement system were chosen 7

Introduction We applied template matching approach for BCG beat detection algorithm to the three different types of BCG measurement system 8

Method The detection method is based on a “template matching” rule evaluated using a correlation function in a local moving-window procedure Beat detection algorithm operates in two stages ―BCG template modeling ―Beat detection by Template Matching 9

Method BCG template modeling 10

Method BCG template modeling ― Bandpass filter  The bandpass filter which has 0.5~20Hz cutoff frequency was applied to remove a respiration signal from raw signal 11

Method BCG template modeling ―The template bases were selected by the following criteria  Clear to identify the I-J-K complexes  Includes at least 10 BCG cycles  Free from respiration effort signal and motion artifact signal 12

Method BCG template modeling ―Segmentation & verification  In the segmentation step, the selected template bases were split to several BCG cycles and each cycle was verified by the expert 13

Method BCG template modeling ―Ensemble Average  And the cycles were normalized between -1 to 1. Finally, BCG template was constructed by an ensemble averaging of the valid BCG cycles centered at J peak points 14

Method Constructed BCG templates : (a) Air-mattress system (b) Loadcell system (c) EMFi-film system 15

Method Beat detection by Template Matching ―Template matching was performed by local moving window function which generates correlation coefficient between the constructed template in previous modeling procedure and BCG signal 16 Template matching illustration

Method 17 Photo Source :

Result Synchronized ECG was measured simultaneously for a convenience of expert’s manual scoring of BCG J peaks 18 Photo Source :

Result Sampling rate: ―Loadcell type BCG system: 200Hz ―Air-mattress type BCG systems:1KHz ―EMFi-film type BCG systems:1KHz 19

Result The template matching approach was exercised in BCG signals to detect the J-peak events Detected J peaks marking with reversed triangle in three BCG systems ― Air-mattress (upper) ― Loadcell (middle) ― EMFi-film (lower) 20

Result The sensitivity is the detection screening probability of the method The positive predictivity value determines the capacity to identify a true event 21

Result We analyzed 10 subjects recorded during resting and sleep. All subjects have normal health condition and signal was acquired in supine position with normal breath. The subjects were recorded for 30 seconds and 5 records were analyzed in each system Result from air-mattress system and the loadcell system shows a high sensitivities and positive predictivity values. 22

Discussion The template matching has an advantage with a simple and fast algorithm, and be able a real-time process Various template types can be useful in different measurement systems and it is possible to register and classify multi-templates with patient disease condition 23

Discussion Fixed template is not a universal set with wide variation of BCG shape according to the change of measurement situations In the further studies, we will apply the template matching approach to classification of the beats with a cardiac disease and an automatic template updating method to overcome the limitation of the fixed template 24

25 Thanks for your attention