BALLISTOCARDIOGRAPHY DEVICE AND METHOD

The mixed filter system in the ballistocardiography device addresses DC drift and noise issues, enabling precise cardiac and respiratory mapping with reduced saturation and faster stabilization, suitable for real-time monitoring.

FR3108026B1Active Publication Date: 2024-11-29FEALING +2
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
FR2020002547
Authority / Receiving Office
FR · FR
Patent Type
Patents
Current Assignee / Owner
Filing Date
2020-03-16
Publication Date
2024-11-29
Estimated Expiration
2040-03-16

AI Technical Summary

Technical Problem

Existing ballistocardiography devices are sensitive to DC drift, noise, and saturation due to patient movement, leading to reduced monitoring time and inaccurate cardiac and respiratory rate detection, especially in noisy environments.

Method used

A ballistocardiography device with a mixed filter system comprising an analog-digital converter, digital filter, and digital-analog converter, allowing for rapid adjustment of digital filter coefficients, enabling precise cardiac mapping and noise reduction.

Benefits of technology

The device provides high-resolution cardiac and respiratory mapping with reduced saturation and faster stabilization times, suitable for real-time monitoring in noisy environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 00000019_0000
    Figure 00000019_0000
  • Figure 00000019_0001
    Figure 00000019_0001
  • Figure 00000020_0000
    Figure 00000020_0000
Patent Text Reader

Abstract

TITLE: BALLISTOCARDIOGRAPHY DEVICE AND METHOD The ballistocardiography device comprises: - a support for a person, - a plurality of pressure or acceleration sensors, each providing an analog signal representative of a pressure or acceleration measured at a point on the support or on the person's body, - a multiplexer (41) configured to receive a plurality of analog signals from sensors and to provide, at the output, a signal successively representative of the input signals, and - an operational amplifier (52) one input (57) of which is connected to the output of the multiplexer and equipped, at its other input, with a mixed filter comprising an analog-to-digital converter (53) converting the signal output from the multiplexer into digital data.a digital filter (54) acting on the digital data and a digital-to-analog converter (55) converting the filtered digital data into an analog signal injected onto this other input (58) of the operational amplifier. Figure for the abbreviation: Figure 2,
Need to check novelty before this filing date? Find Prior Art

Description

Title of the invention: BALLISTOCARDIOGRAPHY DEVICE AND METHOD TECHNICAL FIELD OF THE INVENTION

[0001] The present invention relates to a device and a method for ballistocardiography. It applies, in particular, to ballistocardiography, that is to say to the non-intrusive measurement of mechanical cardiac activity. STATE OF THE ART

[0002] Ballistocardiography (BCG), a non-contact heart rate monitoring method, can measure vital signs and reduce patient pain and discomfort in monitoring. It uses an accelerometer placed on the side of the mattress. Since the signal amplitude is very low, the front-end signal conditioning electronics need a high amplification gain and are therefore very sensitive to DC drift and background noise.

[0003] However, known analog filters are slow to stabilize, and the patient plus the bed constitute a mass-spring mechanical system that generates noise at its resonant frequency. Therefore, the BCG signal is likely to saturate for several seconds whenever the patient moves.

[0004] In the prior art, analog amplification is required to achieve good detection performance. It is therefore very sensitive to the drift of the DC component that occurs when the patient's position changes, for example from lying on the stomach to lying on the back, and must be filtered. However, for measuring respiratory activity whose low-frequency components are around 0.1 Hz, conventional high-pass filters are slow to stabilize and the BCG signal can saturate for several tens of seconds. Adding non-linear components, for example switching diodes in the amplifier's feedback loop, can help reduce the amplifier's stabilization time, i.e. the time before saturation ends, to only a few seconds.In addition, the patient and their bed can be considered as a mass-spring system that generates noise at its resonant frequency. Ambient noises, for example, nurses walking, ventilation devices, or during transport in the pediatric intensive care unit, can also occur. For the same saturation reasons, these repetitive noises must be filtered out, otherwise the monitoring time may be reduced. PRESENTATION OF THE INVENTION

[0005] The present invention aims to remedy all or part of these drawbacks.

[0006] To this end, according to a first aspect, the present invention relates to a ballistocardiography device which comprises: - a support for a person, - a plurality of pressure or acceleration sensors each providing an analog signal representative of a pressure or an acceleration measured at a point of the support or on the body of the person, - a multiplexer configured to receive a plurality of analog signals from sensors and to provide, at the output, a signal successively representative of the input signals and - an amplifier, one input of which is connected to the output of the multiplexer and provided, on its other input, with a mixed filter comprising an analog-to-digital converter converting the signal coming out of the multiplexer into digital data, a digital filter acting on the digital data and a digital-to-analog converter converting the filtered digital data into an analog signal injected into this other input of the amplifier.

[0007] Thanks to these provisions, the state and coefficients of the digital filter can be modified quickly enough for each sensor to be associated, successively, with a specific digital filter. The device can thus implement hundreds, even thousands of sensors and carry out a more detailed and less expensive analysis of the collected data. In addition, the amplification rate can be higher because the continuous component can be better filtered. Each of the signals to be processed can therefore be of better quality and further improve the analysis carried out. Finally, each digital filter used can have a higher order than analog filters. The invention thus makes it possible to carry out more precise cardiac mapping than what was possible with known ballistocardiography devices.

[0008] In embodiments, the device comprises, for each sensor, a memory of the digital data successively outputting from the digital-analog converter when the signal outputting from this sensor is supplied at the output of the multiplexer, and the filter comprises a processor which processes the successively stored data to apply a filtering of this data and supply digital data which is converted by the digital-analog converter.

[0009] Thanks to these arrangements, the signals from the different sensors are processed independently but taking into account their evolution to eliminate the drift of the continuous component and stabilize the filtering applied.

[0010] In embodiments, the digital filter is a cascade of biquadratic filters, otherwise known as cascaded biquadratic filters.

[0011] It is an efficient filter, very easily implemented in embedded systems such as microcontrollers.

[0012] In embodiments, the sampling frequency of the analog-to-digital converter is equal to the product of the number of sensors times a number between 128 and 512.

[0013] Thanks to these arrangements, heart and breathing rates are finely detected.

[0014] In embodiments, the plurality of sensors constitute at least one pressure sensor array.

[0015] Thanks to these arrangements, hundreds, even thousands, of sensors are implemented and the ballistocardiographic or cardio-respiratory mapping has a high resolution.

[0016] In embodiments, the gain of the amplifier is greater than 10.

[0017] The vibrations detected are thus of very low amplitude, which increases the precision of the device.

[0018] According to a second aspect, the present invention aims at an application of a device which is the subject of the invention to the monitoring of a patient during a surgical operation.

[0019] According to a third aspect, the present invention aims at an application of a device which is the subject of the invention to the visualization of the pulse wave and the contraction of the ventricles.

[0020] This application allows for a simplified scan of vital functions, for example visualization of pulse rate and ventricular contraction, at low cost and in real time during an operation or while monitoring a patient.

[0021] According to a fourth aspect, the present invention aims at a ballistocardiography method implementing a device which is the subject of the present invention, which comprises the following steps: - a plurality of pressure or acceleration captures at different points of a person's support or on the person's body, providing, for each capture point, an analog signal representative of a pressure or acceleration measured at a point of the support or on the person's body,

[0022] - a multiplexing of the plurality of analog signals and - an amplification of the multiplexed signal from which a signal filtered by a mixed filter comprising an analog-to-digital converter converting the multiplexed signal into digital data, a digital filter acting on the digital data and a digital-to-analog converter converting the filtered digital data into an analog signal is subtracted.

[0023] In embodiments, the method which is the subject of the invention further comprises a step of estimating the continuous component and a step of initializing the filter with the estimated continuous component.

[0024] Thanks to these provisions, the stabilization time of the filtered signal is very reduced.

[0025] In embodiments, the method which is the subject of the invention further comprises: - a step of detecting a vibration in the amplified signal, - a step of segmenting the vibration signal, - a transformation step, in the frequency domain, of segments of the signal, - a step of obtaining a maximum of the signal, in the frequency domain, - a step of configuring a band-stop filter comprising the maximum obtained and - a step of filtering the amplified signal with the band-stop filter.

[0026] Thanks to these provisions, the noise, for example generated by the patient's movements, is extracted from the processed signal and visualized.

[0027] The aims, advantages and particular characteristics of the method and applications which are the subject of the present invention being similar to those of the device which is the subject of the present invention, they are not recalled here. BRIEF DESCRIPTION OF THE FIGURES

[0028] Other advantages, aims and particular characteristics of the invention will emerge from the following non-limiting description of at least one particular embodiment of the device and method which are the subject of the present invention, with reference to the appended drawings, in which:

[0029] [Fig.l] schematically represents a particular embodiment of the device which is the subject of the invention,

[0030] [Fig.2] schematically represents a front-end electronic circuit for conditioning the signal from sensors,

[0031] [Fig.3] schematically represents an example of a fourth-order IIR filter,

[0032] [Fig.4] represents a distribution of the poles of an IIR filter quantized on six bits,

[0033] [Fig.5] represents a quantization of a pole of an IIR filter quantized on 64 bits to 32 bits,

[0034] [Fig.6] represents, in the form of a frequency function curve, an effect of the quantization and rectification of the bandwidth of an IIR filter,

[0035] [Fig.7] shows frequency responses of the low-pass equivalent of the designed filters and the typical analog RC filter,

[0036] [Fig.8] represents steps of a pseudo-algorithm for identifying and filtering the noise frequency,

[0037] [Fig.9] is an illustration of the suppression of the DC component drift with an amplification of 21 dB,

[0038] [Fig. 10] represents a comparison of the frequency responses for a gain of 21 dB, IIR quantified with respect to experimental points,

[0039] [Fig. 11] represents an effect of the initialization of the filter state on the stabilization time,

[0040] [Fig. 12] represents a stabilization time for different initializations of the state of the filter with an amplification of 21 dB,

[0041] [Fig. 13] shows an example of a BCG pulse and its spectrum of unilateral frequency calculated by a microcontroller,

[0042] [Fig. 14] represents, in amplitudes, the frequency response of different filters,

[0043] [Fig. 15] represents, in phases, the frequency response of different filters,

[0044] [Fig. 16] represents the time response of the filter and the amplifier in the FIR configuration,

[0045] [Fig. 17] represents the time response of the filter and the amplifier in the IIR configuration one section,

[0046] [Fig. 18] represents the time response of the filter and the amplifier in the five-section IIR configuration and

[0047] [Fig. 19] represents the time response of the filter and the amplifier in the FIR configuration with filter state initialization. Description of the embodiments

[0048] The present description is given without limitation, each characteristic of an embodiment being able to be combined with any other characteristic of any other embodiment in an advantageous manner.

[0049] It should be noted from now on that the figures are not to scale.

[0050] The terms processor and microcontroller are used interchangeably in the description. In [Fig.l], we observe a schematic view of a first embodiment 30 of a device which is the subject of the present invention.

[0051] In [Fig.l], we observe a support 31 for a person (not shown), sensors 32 to 35 disseminated in or on the support 31, a front-end electronic circuit 36 ​​for conditioning and amplifying the signals from the sensors 32 to 35, an analog-to-digital converter 37 of the data leaving the circuit 36, a microcontroller 39 and a computer 38 for processing the data.

[0052] The support 31 is, for example, a bed, a mattress, a cushion or a seat. The sensors 32 to 35 are, for example, pressure sensors, possibly in the form of matrices as described below, or accelerometers. A particular embodiment of the circuit 36 ​​is shown in [Fig. 2]. The data processing computer 38 produces, for example, a cardiac or pulmonary map of the person, based on the digital signals coming out of the analog-to-digital converter 37, according to known algorithms.

[0053] The front-end circuit 36 ​​identifies the resonant frequency of the system, filters out the DC component drift and damping noise, and strongly amplifies the signal without saturation, or with a reduced saturation time, after the patient's movement. The front-end circuit 36 ​​comprises, in the embodiment illustrated in [Fig.2], a multiplexer 41 of the analog signals coming from the sensors 32 to 35, followed by an instrumentation amplifier 52 whose output is connected to the analog-digital converter 37. The amplifier 52 has a very high gain, for example greater than 10, or even 100. The analog-digital converter 37 is preceded by an anti-aliasing filter integrated in the analog-digital converter 37.

[0054] The multiplexed signal from the multiplexer 41 is supplied to one input (here the positive input 57) of the amplifier 52 and to an analog-to-digital converter 53. A digital filter 54 filters the digitized signal and a digital-to-analog converter 55 converts the digitally filtered signal and delivers it to the other input (here the negative input 58) of the amplifier 52. A resistor 56 controls the gain of the amplifier 52, between the inputs of the operational amplifier 52.

[0055] The digital signal processor (or DSP for Digital Signal Processor), which produces the digital filter 54 is a 32-bit STM32L476 (registered trademark), mounted on a Nucleo-L476RG card (registered trademark), whose analog-to-digital converter (or ADC for Analog to Digital Converter) and digital-to-analog converter (or DAC for Digital to Analog Converter) channels have a resolution of 12 bits and a sampling frequency fs of between 128 and 512 Hz, for example 256 Hz per sensor.

[0056] The frequency of the multiplexer is therefore N xfs, of at least 130 kHz for

[0057] achieve the minimum spatio-temporal resolution for 1024 sensors (fs = 128 Hz and N = 1024 sensors) and up to 2 MHz for the maximum spatio-temporal resolution (fs = 512 Hz and N = 4096 sensors). The quantizer resolution is 8 to 16 bits, typically 12 bits.

[0058] The microcontroller 39 demultiplexes the filtered signals at the output of the analog-digital converter 37 and conditions these signals into a matrix and then sends them to the computer 38. The computer 38 then displays the matrices in the form of a video and then possibly compresses these matrices. The ARM MBED operating system (registered trademark) was used with the CMSIS DSP library (registered trademark). A universal serial port (or USB for Universal Serial Bus) is used for serial communication with a computer and powering the circuit.

[0059] The sensor is a LIS344ALH (registered trademark) accelerometer with a low noise density of 50 ug / ^Hz and a high sensitivity of 0.66 V / g. It is mounted on a STEVAL-MKI015V1 (registered trademark) evaluation board, and the FS, PD, and ST pins are grounded. It is powered by a single 3.3 V supply and measures accelerations of + / - 2 g in this configuration. It is oriented such that the Z axis is in the direction of gravity.

[0060] The mixed front-end signal conditioning circuit 36 ​​has a mixed signal architecture, i.e. a topology where an analog signal is digitized by an analog-to-digital converter ADC, filtered on a signal processing processor DSP and then synthesized by a digital-to-analog converter DAC. Such circuits have the same advantages as digital filters, for example a sharp transition frequency band or a very short settling time, and can be associated with an analog amplifier.

[0061] In the case of BCG, unwanted components of the analog acceleration signal must be filtered before amplification and digitization. The designed architecture, illustrated in [Fig.2], eliminates these unwanted components and amplifies the resulting signal using an instrumentation amplifier.

[0062] The resulting signal is expressed by equation 1, where G represents the gain of the amplifier and f represents the digital filter.

[0063] Vout = G. [Vin-f (Vin) ] eq. 1

[0064] When synthesizing the return signal, saturation is implemented in order to comply with equation 2.

[0065] 0 <G.[ Vin-f (Vin) ] + Vref < 3.3 équ. 2

[0066] Design and implementation of digital filters: in order to eliminate unwanted components of the signal, two types of digital linear filters have been implemented in embodiments of the device which is the subject of the invention. These are finite impulse response (FIR) filters and second-order infinite impulse response (IIR) filters, shown in [Fig.3] and commonly called cascade of biquad filters. The FIR and biquad filters are respectively defined by equation 3 and equation 4, where n is the sampling index, x and y are the input and output signals, N is the number of coefficients, {a_n} and {b_n} are the feedback and reaction coefficients.

[0067] .yf^] =èojr[n] + ... vb^^nN + l]é0u- 3

[0068] y[ / z] =b(yx[n] + b^[n-1] + b2.x[n-2] -ia^tn-1] +a2.y[n-2] equation 4

[0069] By convention, a0 is set to 1.

[0070] In [Fig.3], we observe an example of a 4th order IIR filter 60, cascade of two biquad filters 61 and 62, as implemented by the CMSIS DSP library.

[0071] The filter coefficients are calculated using the Python library scipy.signal (registered trademarks) on a 64-bit computer. They must be quantized to 32-bit coefficients before being implemented on the DSP signal processor, which can generate errors, especially for IIR filters with highly constrained filters and a very low cutoff frequency. For example, Figures 4 and 5 illustrate the pole quantization for a second-order Butterworth filter with a cutoff frequency of 0.05 Hz and a sampling rate of 256 Hz.

[0072] In [Fig.4] the distribution of the poles 63 quantified under six bits is represented according to a horizontal axis representing their real part and a vertical axis representing their imaginary part.

[0073] In [Fig.5], we observe the quantization of a 64-bit pole 64, before quantization, into a 32-bit pole 65, after quantization. Each of these poles is represented along a horizontal axis representing its real part and a vertical axis representing its imaginary part.

[0074] This coefficient quantization is likely to modify the frequency response of the filter, for example the bandwidth gain as illustrated in [Fig.6] using the same filter as that of figures 4 and 5. After quantization, the bandwidth gain was systematically corrected to 0 dB using equation 5.

[0075] ., , £«, eq. 5 *»=

[0076] We observe, in [Fig.6], with a horizontal axis representing the frequencies and two vertical axes representing, at the top, the amplitudes and, at the bottom, the phases, an unquantized signal 70, a quantized signal 71 and a rectified signal 72. [Fig.6] thus represents the effect of quantization and rectification of the bandwidth.

[0077] The filter stabilization time depends on the initialization. The filter state is usually initialized to zero: the signal is therefore causal. However, the sensor has a rest position that is known or can be estimated. In the case of the accelerometer described elsewhere, it can be assumed that it is lying flat on the mattress which is itself horizontal: the z axis should see an offset value of approximately +1 g, or 2.31 V in this setting. The filter is initialized by considering the signal to be 2.31 V before starting the recording.

[0078] Concerning the suppression of the drift of the DC component, this represents an often undesirable component of the physiological signal. It is a low-frequency component which prohibits high-gain amplification. In ballistocardiography, the displacement of the DC component is a voltage shift dependent on the position of the body and the weight in the bed. For example, when the patient's position changes, the orientation of the accelerometer and the projection of gravity onto the accelerometer axes change. The inventors have observed that this shift value is generally between 0.1 and 0.5 g, for which the amplifier would saturate with G > 3 dB.

[0079] Respiration, whose frequency varies from 0.5 to 1.0 Hz in neonatology and up to 0.1 Hz in geriatrics, can be considered either as a part of the drift of the continuous component, or as a useful component of the signal. In the embodiment described here, the breathing is not filtered, therefore the cutoff frequency of the basic filter is set to 0.05 Hz and f is a low-pass filter (Equation 1).

[0080] In the case of FIR, the window method is applied with a Kaiser window of 40 dB minimum attenuation in the passband and a transition bandwidth of 0.1 Hz. The resulting FIR filter has 5716 coefficients.

[0081] In the case of the FIR, a second-order Butterworth filter, chosen because of its flat maximum response in the passband, is designed and repeated twice in order to accentuate the transition. In the embodiment described herein, the filter chosen is a fourth-order IIR filter, because the FIR filter is computationally expensive for the microcontroller 54.

[0082] [Fig.7] illustrates the responses of the FIR and IIR filters designed in low-pass equivalent. They are compared to a typical first-order resistor-capacitor (RC) filter with the same cutoff frequency.

[0083] In [Fig.7], we observe the frequency responses of the low-pass equivalent of the FIR filters, response 75, IIR, response 76 and of a typical analog RC filter, response 77. Note that the axes are identical to those in [Fig.6], but with a greater extent of the frequency range.

[0084] A gain of 21 dB is set in this embodiment. The high-pass filter is finally equivalent to the following numerator and denominators in Table 1. It is defined by Equation 4, Equation 5 and Equation 6.

[0085] ywirf = g(y^)equ.6

[0086] «» = eq. 7

[0087] »'„ = G.(«„-Wequ.8

[0088] [Tableauxl] aO 1 al -1.9982646 a2 0.99826604 bO 11.183295 bl -22.3472 b2 11.163904

[0089] Numerators aO, al and a2 and denominators bO, bl, b2 of the equivalent high-pass filter.

[0090] Regarding the identification and filtering of noise frequencies, the noise frequency is identified and filtered using the integrated algorithm 80 illustrated in [Fig.8].

[0091] In [Fig.8], we observe an input 81 of the amplified signal, a vibration detection step 82, a vibration segmentation step 83, a transformation step Fast Fourier Transform (or FFT) 84, a maximum extraction step 85 which provides the noise frequency 86, a notch filter selection step 87, a feedback filter update step 88 and a filtered signal output step 89.

[0092] The vibration detection step 82 is parameterized by a voltage limit value, hereinafter referred to as “threshold”, beyond which the vibration is detected. The vibration begins when this limit value is exceeded, by the amplified signal, and has a duration considered to be fixed.

[0093] The power spectral density of the segmented vibration is calculated by the microcontroller. Its maximum is located at the noise frequency, so it is outside the frequency range of the BCG, otherwise the heartbeats of the BCG signal would be distorted or filtered.

[0094] Several notch filters, with quality factors Q=0.707, were preliminary calculated using the scipy.signal library outside the microcontroller. The notch filter whose frequency is closest to the noise frequency is selected, and added to the basic IIR filter.

[0095] In this section, the results of the experiment are detailed: the filtering of the DC component drift, the comparison of the stabilization time and the identification and filtering of the noise are analyzed. The raw signals presented are not post-processed.

[0096] Filtering of the continuous component

[0097] The IIR filter successfully filters out the DC component, and this is eliminated by the amplifier. In the steady state, the maximum peak-to-peak amplitude of the pulses is ten times higher at the output than at the input. [Fig.9] is an illustration of the signal before and after the amplifier with a pulse amplitude below the noise level, with an amplification of 21 dB.

[0098] In [Fig.9], we observe a curve 90 at the sensor output and a curve 91 of amplified acceleration. The horizontal axis represents the time, in seconds and the vertical axis represents the voltage. We observe that a pulse, with a frequency of approximately 60 hertz is extracted from the initial noisy signal, despite the defects of the initial signal.

[0099] This filter makes it possible to generate vibrations at different constant frequencies and to compare the accelerations measured before and after filtering and amplification by their quadratic mean values. The experimentally measured frequency response 95 is compared to the theoretical response 96 of this filter, in [Fig. 10], for the quantized IIR filter, with a gain of 21dB.

[0100] Comparison of stabilization times

[0101] The signals were recorded at different initialization levels. [Fig.l 1] is a superposition of two signals at different times, with an initialization offset from 0 V, curve 101, and 2.31 V, curve 102. We observe that when the filter state is initialized to 0 V, the amplifier can saturate for tens of seconds, which is not the case if the offset is well estimated.

[0102] The stabilization time, on the ordinate, depends on the initialization of the shift, on the abscissa, as illustrated by curve 106 in [Fig.12], with an amplification of 21 dB.

[0103] Noise identification

[0104] The threshold and duration were set to 1.5% and 0.25 s of the initial DC component drift, respectively. The noise was generated by pulsed vibrations, i.e., the simulated BCG. The frequency 112 of the noise 111 determined by the microcontroller is about 68 Hz, as can be seen in [Fig. 13]. This is the resonant frequency of the damping system.

[0105] A notch filter was then selected by the algorithm illustrated in [Fig.8] to filter the identified noise frequency.

[0106] In conclusion, the present invention implements a novel signal conditioning front-end circuit that identifies the system resonant frequency, filters DC component drift and damping noise, and amplifies the BCG signal without saturation after patient movement.

[0107] In the above description, the amplification gain is 21 dB to illustrate the principle of the BCG front-end circuit. However, the invention is not limited to this gain level, especially when a specific electronic circuit with an adequate ground plane is used.

[0108] Compared to conventional BCG instruments, this front-end conditioning circuit described above has a shorter settling time and a sharper frequency transition. It opens up possibilities for medical applications, in real time and at low cost.

[0109] The integration of this front-end circuit into a BCG device is described below. It is recalled that the frequencies 0.05 Hz to 25 Hz include cardiac and respiratory phenomena.

[0110] The purpose of this device is to establish a cardio-respiratory map of the body surface by time multiplexing of an analog filter with a digital feedback loop.

[0111] A mixed filter arrangement 36 as described above replaces an analog arrangement in the feedback loop of an operational amplifier, called digital feedback.

[0112] This digital "feedback" makes it possible to increase the constraints of the filter, to initialize it digitally so that it is faster, to attenuate ambient noise whose bandwidth evolves over time and to carry out high-frequency time-division multiplexing of filters with low cut-off frequencies.

[0113] The device is a BCG imaging system which comprises the following elements: - A surface sensor of body pressure, - An acquisition unit with time division multiplexing of a filter and an amplifier, - Digital signal processing software.

[0114] The surface body pressure sensor is usually used to measure the comfort of an individual in a lying or sitting position and / or the risk of pressure sores. It is generally composed of several modules of size 35x40 cm to 50x100 cm, with resolutions ranging from 0.5 to 1 sensor / cm2, or N = 1024 to 4096 sensors per module. Example of Body Pressure Measurement Systems (in French, body pressure measurement systems) from the company Tekscan (registered trademark).

[0115] One acquisition unit is implemented per module, i.e. for more than a thousand pressure sensors.

[0116] The filter, linear or non-linear, can be configured by its bandwidth, transition width, maximum attenuation in the bandwidth and minimum attenuation in the cutoff band. The amplifier has a configurable gain. The filter and the amplifier can be configured in real time, for example to amplify a signal or attenuate noise whose bandwidths change over time.

[0117] Let fs be the signal sampling frequency. To measure cardiac function, the filter bandwidth is set to [1; fs / 2] and the gain is set to 7 to 40 dB (x5 to xlOO). To measure respiratory function, the filter bandwidth is set to [0.1; 1] and the gain is set to 7 to 40 dB. To measure both functions simultaneously, the filter bandwidth is set to [0.5; fs / 2] and the gain is set to 7 to 40 dB. In both cases, the DC component of the signal must be removed so as not to saturate the amplifier (AC coupling).

[0118] The analog-to-digital converter includes an anti-aliasing filter with a cutoff frequency of fs / 2.

[0119] The sampling frequency of the signal fs is between 128 and 512 Hz, the frequency of the multiplexer is therefore N x fs, from at least 130 kHz to achieve the minimum spatio-temporal resolution (fs = 128 Hz and N = 1024 sensors) up to 2 MHz for the maximum spatio-temporal resolution (fs = 512 Hz and N = 4096 sensors).

[0120] Finally the resolution of the quantizer is 8 to 16 bits, typically 12 bits.

[0121] In order to quickly multiplex a filter with such a slow bandwidth, we uses an operational amplifier with a digital feedback loop.

[0122] An advantage of the arrangement illustrated in [Fig.2], in which the filter bandwidth is subtracted from the signal Vin by the amplifier, is that the filter is a digital filter executed on a microcontroller, preceded by an ADC and followed by a DAC. It can be of any order, with finite or infinite impulse response, linear or non-linear, or even adaptive.

[0123] In addition, the state of the filter can be initialized so that the amplifier saturates for a shorter time, as if the capacitor of an analog filter were precharged to its equilibrium value.

[0124] Finally, the state and output of this filter can be stored at each sampling step in a memory: it can therefore be time multiplexed, even if the frequency is much higher than the filter cutoff frequency. In addition, the digital filter in the feedback loop makes it possible to remove the noise frequency bands in order to prevent the latter from saturating the amplifier.

[0125] The digital signal processing software can be embedded in the acquisition card for real-time processing or on a computer for post-processing. It interpolates the measurements to digitally increase the image resolution and allows the clinician to visualize cardiac and respiratory functions, and quantifies these functions (heart and respiratory rates, pulse rate trajectory, etc.).

[0126] The advantages of this device are numerous. In a hospital setting, it allows cardiac and respiratory function tests to be carried out simply and quickly, by being integrated directly into the bed. At home or in non-medical contexts, it allows short and long-term monitoring of vital data (bed, chair, seat, clothing).

[0127] It has the same advantages as known type ballistocardiography systems: - it is less sensitive to layers of clothing and sheets compared to other measuring systems, for example by Doppler ultrasound or CCD camera. - it allows to measure cardiac and respiratory functions simultaneously.

[0128] It has other advantages than known ballistocardiography systems: - it performs a surface and high-resolution measurement (>1000 points), - it is more robust thanks to active noise attenuation and therefore more easily usable in noisy environments, - it is faster thanks to the initialization of the digital filter.

[0129] In addition, its acquisition unit is simple, compact and low cost.

[0130] In embodiments, the instrumentation amplifier used is an AD623AN (registered trademark) whose gain is adjustable using a resistor. In experiments, two gains are used: a unit gain (no resistor, open circuit) and an xl0 gain (resistance of approximately ten kΩ). The results presented below are validated for an xl00 gain with a stabilized power supply and a circuit having an adequate ground plane.

[0131] Figures 8 and following describe the processing of signals from accelerometers. Apart from the units and values, these signals are similar to signals from pressure sensors, with in particular a continuous component, noise and physiological signals to be extracted.

[0132] In implementation examples, the accelerometer output is filtered on the Nucléo card (registered trademark), using the MBED-DSP library and in particular the arm_fir_f32() and arm_biquad_cascade_dfl_f32() functions.

[0133] Filter design

[0134] Two types of filters are studied: - FIR filters - IIR filters.

[0135] They are designed with the Python scipy.signal library.

[0136] The FIR filter is of order numTaps-1 and consists of numTaps coefficients {bi].

[0137] y[n] = b[0] * x[n] + b[l] * x[nl] + b[2] * x[n-2] +...+ b[numTaps-l] *x[n-numTaps + 1]

[0138] The FIR filter is calculated by the window method, with an optimal Kaiser window. This is determined with the constraint of an attenuation of 40 dB in the passband and a transition of the order of 0.1 Hz (Nyquist frequency 128 Hz):

[0139] numtaps,beta = scipy.signal.kaiserord(40,0.1 / 128)

[0140] The number of coefficients in this optimal window amounts to 5716. With the bandwidth constraints (cutoff frequency at 0.05 Hz) we obtain: b = scipy.signal.firwin(numtaps=numtaps,cutoff=0.05,window=('kaiser',bet a),pass_zero=True, fs=256).

[0141] The biquadratic IIR filter, first-order Butterworth, is of order two and contains a denominator. For equivalent transition width and attenuation, the order of the IIR filter is much lower than for the previous FIR filter but it can be less stable.

[0142] y[n] = bO * x[n] + bl * x[nl] + b2 * x[n-2] + al * y[nl] + a2 * y[n-2]

[0143] It is obtained by the following functions (coefficients in appendix?): b,a = butter(2,0.05 / (fs / 2)); pCoeffs = concatenate((b,-a[l:]));

[0144] It is noted that the first-order Butterworth filter is interesting to study because it is the digital equivalent of the analog filter produced in electronic AC coupling circuits (in alternating current): a capacitor followed by a resistor, with possibly an operational amplifier to resolve impedance issues or to add gain.

[0145] In a cascaded biquadratic IIR filter, sections of the previous IIR filter are placed in series.

[0146] As discussed above, the 64-bit coefficients calculated on the computer must, in embodiments, be quantized into 32 bits, and the filter response varies enormously because the stability margins are small. As a result, some biquadratic filters designed under the 64-bit architecture are unstable under 32 bits.

[0147] It is noted that the number of coefficients is much lower than for an FIR filter, so the number of operations is much lower, which is favorable to the signal processing speed. The structure of the filter can also be improved to further reduce the number of operations (type II structure) or to reduce the influence of the quantization of the coefficients (Gold & Rader structure).

[0148] Figures 14 and 15 represent, in amplitudes and phases, the frequency responses of different filters. Curves 113 and 116 represent the response of an FIR filter. Curves 114 and 117 represent the response of a second-order IIR filter. Curves 115 and 118 represent the response of a tenth-order IIR filter. We can see the advantage of using digital filters (more abrupt transition, greater attenuation). The response of the second-order IIR filter is equivalent to that of a first-order analog low-pass filter.

[0149] It is noted that, in other embodiments, other filters are chosen, for example non-linear filters (median filter, wavelets, etc.) and also for uses other than AC coupling.

[0150] Concerning saturation, the AD623 amplifier is powered at 0 - 5 V so its output voltage, in case of saturation, can exceed the maximum voltage of the ADC. It is necessary either to add hardware saturation on the amplifier output, or software saturation on the DAC output. This second method has advantages in terms of simplicity of implementation: float expectedOutput = (rawSignal - filteredSignal)*analogGain+Vref; if (expectedOutput > 3.3){ feedback. write((ra wSignal+( V ref- 3.3 ) / analogGain) / 3.3 ); } else if (expectedOutput<0){ feedback. write((rawSignal+Vref / analogGain) / 3.3); } else { feedback. write(filteredSignal / 3.3); }

[0151] For the saturation to be correct, it is necessary to precisely measure the voltage Vref as well as the gain: with a voltmeter for the first (1.69 V) and with the resistance value for the second (11.9 kQ or a gain x9.4).

[0152] The state of the filter is usually initialized to zero, that is, all values ​​x[n] and y[n] for which n<0 are zero. The signal is then causal.

[0153] The sensor has a rest position which is known or can be estimated. For an accelerometer in ballistocardiography, it is assumed that it is placed flat on the mattress which is itself horizontal: the z axis therefore sees +lg in DC value. According to the data sheet of the LIS344ALH accelerometer the value, the output DC value of the accelerometer is therefore 2.31 V. We can initialize the state of the filter by imposing that the values ​​x[n] and y[n] are worth 2.31 V for numTaps <n<0.

[0154] For other sensors, it is also possible to create a first fast low-pass filter (higher cut-off frequency) to estimate the DC value of the accelerometer and fill the second long filter with these values.

[0155] Figures 16 to 19 show the time response of the filter and the amplifier in each of the configurations stated above:

[0156] [Fig. 16] represents the 121 response of an FIR filter. We can clearly see the amplification of the pulses (they are hidden in the noise of the raw signal, below the resolution limits of the ADC) but the filter takes a certain time to charge, which is inherent in the fact that the cut-off frequency is very low. This is also true for the analog and amplified filter when the amplifier gain is increased.

[0157] [Fig. 17] represents the 122 response of a 1-section IIR filter. This filter has a behavior identical to that of a first-order analog high-pass circuit.

[0158] [Fig. 18] represents the 123 response of a 5-section IIR filter. The time scale is much longer because there are more coefficients. There is not necessarily much point in increasing the order of the IIR filter.

[0159] [Fig. 19] represents the response 124 of an FIR filter with forced initialization. Logically, an absence of saturation is observed when the FIR filter is correctly initialized. This is a way of "preloading" the filter. A slight oscillation appears, which corresponds to the fact that the value of the DC component provided for the initialization of the filter is slightly different from the value of the actual DC component of the accelerometer.

[0160] We have the same result of absence of saturation when the filter is correctly initialized with the IIR filters.

[0161] The present invention applies, in particular: - monitoring patients during surgery and / or - visualizing vital functions, e.g. visualizing the pulse wave and ventricular contraction.

Claims

Claims

1. Ballistocardiography device (30), characterized in that it comprises: - a support (31) for a person, - a plurality of pressure or acceleration sensors (32 to 35) each providing an analog signal representative of a pressure or acceleration measured at a point on the support or on the person's body, - a multiplexer (41) configured to receive a plurality of analog signals from sensors and to provide, at output, a signal successively representative of the input signals and - an operational amplifier (52) one input (57) of which is connected to the output of the multiplexer and provided, at its other input, with a mixed filter comprising an analog-digital converter (53) converting the signal leaving the multiplexer into digital data,a digital filter (54) acting on the digital data and a digital-to-analog converter (55) converting the filtered digital data into an analog signal injected on this other input (58) of the operational amplifier.,

2. Device (30) according to claim 1, wherein the digital filter (54) is a cascade (60) of biquadratic filters (61, 62).

3. Device (30) according to one of claims 1 or 2, in which the sampling frequency of the analog-digital converter (53) is equal to the product of the number of sensors by a number between 128 and 512.

4. Device (30) according to one of claims 1 to 3, in which the plurality of sensors (32 to 35) constitute at least one matrix of pressure sensors.

5. Device (30) according to one of claims 1 to 4, in which the gain of the amplifier (52) is greater than 10.

6. Application of a device (30) according to one of claims 1 to 5 to the visualization of pulse rate and contraction of the ventricles.

7. Ballistocardiography method implementing a device (30) according to one of claims 1 to 5, which comprises the following steps: - a plurality of pressure or acceleration captures at different points (32 to 35) of a support (31) of a person or on the body of the person, providing, for each capture point, an analog signal representative of a pressure or an acceleration measured at a point of the support or on the body of the person, - a multiplexing of the plurality of analog signals and - an amplification of the multiplexed signal from which is subtracted a signal filtered by a mixed filter comprising an analog-digital converter (53) converting the multiplexed signal into digital data, a digital filter (54) acting on the digital data and a digital-analog converter (55) converting the filtered digital data into an analog signal.

8. A method according to claim 7, which further comprises a step of estimating the DC component and a step of initializing the filter (54) with the estimated DC component.

9. Method according to one of claims 7 or 8, which further comprises: - a step of detecting a vibration in the amplified signal, - a step of segmenting the vibration signal, - a step of transforming, in the frequency domain, segments of the signal, - a step of obtaining a maximum of the signal, in the frequency domain, - a step of configuring a band-stop filter comprising the maximum obtained and - a step of filtering the amplified signal with the band-stop filter.