Towards a heart disease diagnosing system based on force sensitive chair's measurement, biorthogonal wavelets and neural networks
Introduction
The heart disease diagnosing (HDD) system uses a force sensitive electromechanical film (EMFi™) sensor installed under the upholstery of a chair. The sensor generates a movement related signal consisting of components attributable to the ballistocardiogram (BCG), respiration, and body movement. EMFi™ sensor film is an elastic electret material consisting of three distinct layers: two smooth and homogeneous surface layers, and a thicker mid section full of flat air voids separated by leaf-like polypropylene layers. External force supplied to the film surface changes the thickness of the air voids and causes charges residing on the polypropylene/void interfaces to move with respect to each other. As a result, a charge proportional to the force (pressure) applied to the film is generated to the film electrodes (Lekkala and Paajanen, 1999). BCG which typically will trail the ECG by about 0.1–0.3 s is an interesting Bio-measurement for diagnosing, monitoring and managing myocardial disorders related to heart diseases. The disorders are characterized by sudden recurrent and transient disturbances of myocardial function and/or mechanical movements of the heart. The presence of any abnormal patterns in BCG points towards an abnormal heart condition. However, sometimes for normal subjects, in special situations, there are some similar patterns compared to abnormal heart disease related BCG patterns. During the past several years, a large number of biomedical signal processing methods have been developed, including single and multi channel template matching, principle component analysis, amplitude separation, Fourier analysis, linear filtering, autoregressive modeling, neural networks, and maximum likelihood. In the field of BCG signal processing also some methods have been developed and some of them are listed in Yu et al. (1996). Most of the existing methods perform very well while the problem of Motion artifacts, and BCG waveform's latency as well as non-linear disturbances such as electrical drifting of electronic devices or another noise sources are not considered. However, methods that do not deal with such an important issue may potentially give us untrue information about the patient. Other limitations of the existing techniques concern their degree of success in the cases of special situations of subject such as stress, their ease of hardware and/or software implementation, their portability across platforms, and their suitability for real-time processing. To overcome these problems, we used some high-resolution methods in our developed systems which are explained in next sections.
Section snippets
The HDD system
In this section, we will explain briefly about our developed system. We have designed the HDD system based on force sensitive chair's measurement developed in the ProHeMon project (Koivuluoma et al., 2001, Koivuluoma et al., 2004; Junnila et al., 2004) (see Fig. 1). This system consists of some parts shown in a flow chart, Fig. 2. We can categorize devices used for making interaction between Man and Machine (subject→machine→operator/Medical Doctor) to three major parts: measurement, signal
Intelligent signal processing methods
In this section, we will explain briefly the methods we used to process measured BCG data. To recognize the most important BCG features and to decrease information redundancy, the biorthonormal wavelet transform is used. Then these features are presented to an ANN for classification. To evaluate our system performance, we used six subjects belonging to three categories: young healthy, old healthy and old men with a past infarct in their heart. As can be seen in Fig. 5, our suggested intelligent
Results
To demonstrate performance of our approach and for comparing results, we used MLP Neural Networks with four inputs, tanh() to simulate non-linearity of neurons, learning rate for all layers, two hidden layer (15 and 10 neurons, respectively), and three inputs for classifying 30 subjects to three categories: young healthy students with age between 20 and 30 years (10 subjects), old healthy men with age between 50 and 70 years (10 subjects) and ten old male subjects (50 and 70 years old)
Discussion
The proposed BCG signal analysis system consists of a sensitive EMFi™ film based movement sensor, amplifiers and ADC, wavelet-based feature extraction and ANN classification of the features to three classes. Here the EMFi™ sensor has been fitted to an office chair but it could be installed to chairs for homes or even to cars. The advantage of BCG analysis to ECG analysis is that no electrodes are needed to be attached to the subject. Although the present version uses ECG to extract the BCG
Acknowledgments
This study was financially supported by the Academy of Finland, Proactive Information Technology Program 2002–2005, and the Finnish centre of Excellence Program 2000–2005. The authors warmly thank Mrs. Pirjo Järventausta and Ms. Marjaana Ylhänen for carrying out the measurements, and all the subjects for their participation.
References (10)
- Akhbardeh, A., Erfanian, A., 2001. Eye tracking user interface using EOG signal and neuro-fuzzy systems for...
- Akhbardeh, A., Vahabian, A., Farrokhi, M., 2004. EEG features extraction using neuro-fuzzy systems and shift-invariant...
- et al.
Introduction to Wavelets and Wavelet Transforms
(1998) Neural Networks, a Comprehensive Foundation
(1998)- Junnila, S., Koivistoinen, T., Kööbi, T., Niitylahti, J., Värri, A., 2004. A simple method for measuring and recording...
Cited by (16)
Machine learning approach for motion artifact detection in ballistocardiogram signals
2020, PervasiveHealth: Pervasive Computing Technologies for HealthcareA chair based ballistocardiogram time interval measurement with cardiovascular provocations
2018, Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society, EMBSA survey on Ballistocardiogram to study the mechanical activity of heart
2017, Proceedings of the 2017 IEEE International Conference on Communication and Signal Processing, ICCSP 2017Non-invasive BCG monitoring for non-traditional settings
2016, Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society, EMBSEvaluating the force of contraction of heart using ballistocardiogram
2016, 2016 3rd MEC International Conference on Big Data and Smart City, ICBDSC 2016Smart chair based on multi heart rate detection system
2015, 2015 IEEE SENSORS - Proceedings
- 1
EMFi is a registered trademark of Emfit Ltd.