JP2879663B2 - Fetal monitoring device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an improvement in a fetal monitoring apparatus, and more particularly to an apparatus for reducing oversight and interruption of recording by devising a signal processing.

[0002]

2. Description of the Related Art A fetal heart rate monitor, which is a main measuring means for constructing a fetal monitoring apparatus, uses a fetal heart rate signal obtained by an ultrasonic Doppler method as a signal source, filters the fetal heart rate signal, and performs detection such as envelope detection. Then, the periodicity is extracted by a short-time autocorrelation from the result of the pre-processing of slightly filtering, and this is converted into an instantaneous heart rate value and displayed or recorded. Based on the success of this technique (1, 2), which was the inventor's old invention, the only electronic means for ensuring the safety of the fetus during delivery is to attach the probe appropriately on the abdominal wall of the mother. Desired information can be extracted in real time, and it has become a clinically practical object that does not require skilled techniques. At present, this method is the de facto standard for this purpose (3).

The method of detection and processing for instantaneous frequency recognition based on such extremely short-period periodic waveform data is a problem very similar to the so-called automatic pitch extraction of voice (4). It seems that this area generally precedes. Also, since information equivalent to the period of each beat and the change of each beat is required, the section length of the original waveform data to be subjected to the correlation processing to obtain the periodicity is calculated from the heart rate measurement result. Feedback is automatically controlled to limit the length to just over one beat (5).

[0004] With this in mind, techniques that use signal data, such as so-called frequency spectrum analysis and MEM, that handle signal data that is sufficiently long to some extent and in which ergodicity can be expected, have never been studied or adopted in this area. Was.
It was proposed (6) that the instantaneous frequency of an extremely short time waveform could be detected in a self-adaptive or retrospective manner even in the frequency spectrum analysis by DFT / FFT. The established adaptive autocorrelation method has yet to be traced. References (7,8,
An introductory commentary on a fetal monitoring device (also referred to as a delivery monitoring device) is given in 11).

In the first place, statistical and estimating methods such as the present method are used instead of a deterministic method of a trigger generation type based on extraction of local feature points of a waveform of an electrocardiographic signal or a heart sound signal (both are shown in the following figure). Despite the fact that the Doppler fetal heartbeat signal has a clear periodicity when heard by the ear, it is composed of multiple components, the envelope of which is complex, and sometimes has an inconsistent time variant. I do. Such quasi-unsteady signals cannot be handled by conventional deterministic methods that detect local features and generate trigger pulses on the premise that the system and signal waveforms are known in advance. Absent.

[0006] However, the present method, which ensures the periodicity of the signal, that is, the heartbeat detection and verification on the signal processing side, has also been pointed out with a number of difficulties or problems. Here, the problem of escape and tracking of the signal source, that is, the fetal heart, and the problem of misrecognition or inability to adapt, which largely depend on the physiological phenomena of both mother and baby, are pointed out as problems on the signal processing side. For the problem of the ability to follow minute fluctuations and large sudden changes in the heart rate, and the problem of the verification of the probability of the calculation results associated with it, for each scene by means of a two-dimensional image processing method on a correlogram or a reciprocal display correlogram. Introducing recent progress such as.

3a, 3b and 3c show a fetal Doppler signal, its envelope, and a bandpass-processed version of the fetal Doppler signal. Attached to correlation (9). One example of the result is shown in FIG. The first of this correlation function image
The maximum (main maximum) is captured, and the period is recognized from the position. As a mathematical property of the thing called autocorrelation, the position of the main maximum speaks for its periodicity regardless of the specific form of the waveform.
However, it should be noted here that the autocorrelation is not used for improving the S / N or recovering the buried signal at all, but only for overcoming the complexity of the waveform and its temporary alteration. That is the point.

In the period recognition based on the autocorrelation, what is required of the source signal is that there is no situation in which the waveform changes between adjacent beats at all, and a moderately high correlation value while changing gradually. It can be said that the filtered envelope of the Doppler fetal signal satisfies this requirement quite well. Further, it is more advantageous to perform a bandpass process as shown in the figure to process a portion containing the most remarkable periodicity information, particularly in the case of a correlator having a small number of bits, rather than subjecting the envelope signal to the direct correlation process. is there. This is a phenomenon common to the case of speech recognition, and the fundamental wave is not necessarily essential for period recognition.

These can be expressed as follows by borrowing some mathematical expressions.

(Equation 1)

By appropriately adjusting the exponential function term in this equation to limit the components of the past correlation value contributing to the result to approximately a little over one beat, the instantaneous period for each beat, and hence the instantaneous heart rate value, is claimed. Extract the information that can be done. FIG. 5 shows an example of the “Doppler / autocorrelation” fetal heart rate chart thus obtained.

[0011]

A summary of the problems of the current system can be found in Morgenstern et al.'S analysis and evaluation of commercial machines (10). Compared to what he points out, as well as other opinions and rules of thumb, there are still roughly three issues as follows.

(1) The ability to follow the beat-to-beat variation, or the fidelity of its expression Since autocorrelation is a kind of statistical processing, attention must be paid to the ability of past data to contribute to the results. While the correlation function between the new beat and the immediately preceding beat is being developed, a change of old and new occurs at the position of the main maximum, and the calculated heart rate value changes to that of the new beat time, but it is truly instantaneous. In order to be a heart rate meter, the correlation function between the old beats at this point must not have a dominant voice, and the management of the data length for the operation is automatically and adaptively managed as described above. Things are necessary. There is also the idea of filtering (make up) the extracted trend data of the heart rate value on the time axis, but this is strongly warned by the clinical medicine side, and seemingly noisy heart rate charts In fact, it is diagnostically valuable information that most accurately describes heart rate variability (unless it is noisy due to poor input signal quality). This point is still misleading in the design technology or design philosophy of equipment products.

(2) The ability to follow large sudden changes or fidelity At first glance, the faithful reproduction of the fine structure of the heart rate chart and the ability to follow high-speed large fluctuations at the time of labor, etc. seem to be contradictory. In fact, it is the same problem. In any case, there is a problem of how to appropriately use the past calculation result as a reference and not use it in the training of the correlation function and the search for its main maximum. In many intelligent processing devices and processing algorithms, a plurality of algorithms having different viewpoints are always run in parallel, and selected and operated by rules such as weighted majority.

(3) Verification of Probability Is the main local maximum and its position known at the present time valid?
The problem of whether aliases or local peaks are misidentified is determined only by the nature of the signal that is statistically known, the calculated heart rate value, and the background history. Machinery and software are far below human wisdom. However, if the retrospective processing described later is practically used, a considerable portion can be remedied.

[0015] These points greatly affect the procedure and method of correlation calculation and how to interpret the main maximum. Morgen
Stern (10) describes a case where it is difficult to track the position of the main local maximum by posting a correlogram in a scene where a waveform change is severe such that it is difficult to develop a correlation function (that is, statistical integration). According to the author's analysis (1, 11), the ability to follow such fine fluctuations of the main maximum and the prevention of losing are mutually contrary. A: Good signal condition, stable heart rate value If only fine fluctuations are observed, the preprocessing parameters are weighted toward the high-frequency selection side so that the function form becomes as sharp as possible so that the position accuracy of the main maximum is obtained at the highest level. b: In a situation where the signal condition is deteriorating or the heart rate value is fluctuating every moment, so as not to lose sight of it globally without regard to the deterioration of its position accuracy due to the fattening of the main maximum,
Weight the preprocessing parameters to the low-frequency emphasis side. c: When the main maxima and the submaximum are temporarily confusing, or when the order of the main maxima is mistaken and there is a risk of erroneously recognizing double or half the measured value, the history of the heart rate fluctuation and the physiological probability Judge based on knowledge.

If these adaptive controls are carried out without fail, most of the changes in the signal status and physiological status can be absorbed throughout the delivery and the heart rate chart can be maintained. However, even if it is not possible to maintain the measurement in a critical situation, it occurs at a considerable frequency. The design philosophy of the conventional apparatus and method puts priority on "don't lie so as to cause misdiagnosis". As a result, if it is confusing, all measurement results are discarded and recording is stopped. However, if such rejections occur frequently, the heart rate chart can no longer be obtained unless it is indecipherable and of low quality (1).
1).

[0017]

Means for solving the problem The above-mentioned heartbeat value extraction and its adaptation control are all competing in the training and interpretation on the premise that the short-time autocorrelation function of equation (1) is settled. Morgenstern (10) explains this with a correlogram in contalographic notation. However, the original correlation function is obtained after the function of the time difference axis is uniquely obtained as the result of weighted integration in the time axis direction as defined in equation (1). There is no way to review the signal. It is known that even in a trigger-type deterministic heart rate meter, if retroactive remedy for erroneous recognition in the immediately preceding past can be remedied, the quality will be significantly improved. Here, if both the forward and backward processes can be performed in a pseudo manner by allowing a slight waiting time, there is a possibility that much of the missing data can be rescued by the autocorrelation method. Rather than constructing a correlogram using the result of integration of the correlation function of Equation (1) as a unit, each product value that is an element of the product sum is used as a participating element (without integration). Consider the expanded correlogram. In other words, when the mathematical expression is borrowed, it becomes a simple form in which the operation of the exponential weighting integration is eliminated from the equation (1), and each pixel is formed by an element term of the sum of products.

[0018]

(Equation 2)

By imposing a two-dimensional local weighting smoothing filter on this, the main maximum can be found and identified. In a scene where it is easy to find the main maximum, the coverage of the two-dimensional filter is appropriately narrow, giving a high-frequency emphasis character and improving the position accuracy. Conversely, in situations where it is not easy, priority is given to not expanding and missing. This filter is a working filter for the image of the correlogram, not a time axis filter for the contraindicated heart rate trend. However, such an extended correlogram takes a form that saves all the element terms of the sum of products, so occupying an enormous amount of memory if temporary storage is not performed for a long time.
Impractical. However, a few seconds to ten and a few seconds are enough for the proposed work here.
In addition, the amount of embedded memory of the embedded processor can sufficiently withstand this.

FIG. 6A shows an example of a correlogram obtained from the envelope of a real fetal Doppler signal, which is applied to gradation display by luminance modulation.
All peaks such as Arias can be observed at a glance. If all the positions of the vertices are plotted, the trend is pseudo-line-drawn as shown in FIG. 6B. When this is "squinted", it is understood that one line (arrow) tracing the correct heartbeat cycle has penetrated therein.

That is, it is a problem of image processing to find and identify the main role of such a correlogram or the trajectory of the vertex thereof in a line drawing image according to the situation of the scene, and to remove the interruption or contamination. This is a relatively easy task that can result in a local two-dimensional low-pass filter or median filter. This is a new and more preferable method proposed by the present invention, and is a solution that uses both the adaptive processing according to the signal situation as described above and the processing in both the forward and backward (retrospective) directions.

The processing in the time difference axis (vertical) direction on the correlogram corresponds to the processing on the frequency axis of the signal with correlation. Further, the processing in the real time axis (horizontal) direction means such a forward / backward (retrospective) combined processing, whereby it is possible to discriminate and eliminate missing or abnormal data of a small section, and to interpolate or supplement. . Retrospective processing violates causality, but requires only a small delay (in this case, for example, 2
３3 seconds) can be implemented without practical problems.

FIG. 6C and FIG. 6D show the line-drawn image, that is, the "line drawing correlogram", in which the vertical axis is represented by a heartbeat frequency (heartbeat value) which is the reciprocal of the heartbeat period. This is "Reciprocal display line drawing correlogram"
Say

FIG. 7A is a fetal heart rate chart obtained by further purifying the reciprocal display linear drawing correlogram described above. FIG. 7B is a fetal heart rate chart obtained by processing the same signal by a conventional method according to common sense. As is clear from the comparison between the two, such processing can relieve oversight, dropout, and disturbance that cannot be remedied by the conventional method, which is clinically preferable.

The contents of the two-dimensional filter are not limited to the low-pass filter and the median filter as described above, and any symmetrical, asymmetrical, linear, or non-linear filter can be adopted as necessary and effective. The degrees of freedom are endless. Therefore, the gist of the present invention is not restricted by the content of the two-dimensional filter itself. Also, the extent of the range to be processed, that is, the selection of the size (dimension) of the filter is within the degree of freedom in implementation.

The contents of the two-dimensional filter are empirically determined according to the quality (eg, S / N) of the signal, the heart rate value at that time, or within a certain range in the immediately preceding past, or depending on the fluctuation status thereof. Or adaptively changing is as beneficial as in the prior art. In other words, when the signal quality is poor or the heart rate is changing suddenly, the size of the filter is increased so that the emphasis is on finding the correlation peak rather than seeking the measurement accuracy and fidelity, and the probability is high. Increase the allowable range of discrimination. Also, in situations where the signal quality is better than a certain level and the heart rate is almost constant, the size of the filter is kept to the minimum necessary with emphasis on accuracy and fidelity so that the situation of minute heart rate variability is faithfully recorded, In addition, the allowable range of the probability determination is narrowed down and maintained as long as appropriate. Furthermore, if there is a drop or disturbance, go back to the temporarily stored correlogram and try various filters and peak detection algorithms on a trial basis, and use the most likely one from the results output by each. I do.

[0027]

The present invention points out and analyzes some problems in fetal heart rate measurement in which Doppler fetal signals are input and the periodicity is observed by the autocorrelation method, which has not yet been completely solved. In addition, a method based on reciprocal display line drawing correlogram was proposed as a new and improved processing procedure. This is an extension of the conventional method, in which the stage of adaptation to the scene is reduced to image processing on this correlogram, and the degree of freedom is increased by allowing both retry and forward-backward processing to be used freely. Thus, many of the heart rate charts in difficult scenes that have been lost in the past can be rescued.

Further, in addition to the method using the correlation function as an element as the correlogram, it has been proposed to use an extended correlogram in which all the element terms of the product sum before integration are registered as pixels. According to this, the degree of freedom of any retrospective processing including the retry of weighting of the integral at the stage of obtaining the correlation function is guaranteed, and a more advanced processing algorithm can be applied.

[0029]

[References]

(1) Takeuchi, Hogaki, "Adaptive Correl
ation Ratemeter, A New Met
hod for Dopplerfetal gear
rate measurement "Ultras
onics 16-3, 127-137.1978 (2) Takeuchi, U.S.A. S. Pat. 3,991,365 (corresponding Japanese patent is Japanese Patent Publication No. 56-7592) (3) "FIGO News, Guideline f
or the useof fetal monoto
ring ”(FIGO subcommitteon
standards in perinatal m
editine, Zurich, March 22-2
8, 1985) Intl J GynaecolObs
tet, 25: 159-167, 1987 (4) There are countless documents on automatic pitch extraction of speech by autocorrelation. In the sense of review, I suggest the following documents and the documents cited therein. D. A. Kr
ubsack and R.S. J. Niederjoh
n, "AnAutocorrelation Pict
h Detector and Voiceing Det
edition with Confidence Me
asures Developed for Nois
e-CorruptedSpeech ", IEEE T
r. Signal Processing, Vol. 3
9, No. 2,319-329 Feb. 1991 (5) In addition to the above (1) and (2), see (7), etc. (6) Yoshimitsu Takazawa, "Precise Calculation Method of Pitch Frequency by Adjusting Time Window Width", Acoustical Society of Japan Society of Acoustics, MA9
2-13, November 1992 (7) Sakamoto, Hogaki, Hara, Takeuchi, "How to capture fetal information,
Fetal heart rate measurement system using real-time autocorrelation "Obstetrics and gynecology treatment 30,595-601,1975" (8)
Industry Research Committee (9) M. Fano, "Short time a
utocorrelation function a
nd power spectrum "JASA22,5
46- (1950) (10) J. Morgenstern et. al. "
CTG-GerateTest '93 "Medizin
ische Einrichtender He
inrich Heine University,
Frauenlink, Dusseldorf,
1994/1995 (11) Takeuchi, "Reliability, Phenomena and Hardware Aspects of Fetal Heart Rate Measurement" in Obstetrics and Gynecology, Vol. 4
3, No. 13, 1777-1784 (Kanehara Publishing, 1
994) (12) Yamada, Makino, Hayashi, "Study on Peak Tracking Procedure in Correlation Method" Proceedings of the Acoustical Society of Japan, 3-8-5, 1991 (1991) October, 553-
554 pages

[Brief description of the drawings]

FIG. 1 shows an example of a fetal electrocardiogram signal. This is the maternal abdominal wall fetal ECG signal.

FIG. 2 shows an example of a fetal heart sound signal.

FIG. 3 shows an example of a Doppler fetal heartbeat signal. a, b, c
Is an oscilloscope waveform at each stage in the process.

FIG. 4 shows an example of an autocorrelation function obtained from a Doppler fetal heartbeat signal.

FIG. 5 shows an example of a fetal heart rate diagram obtained by a conventional method from a Doppler fetal heart rate signal.

FIG. 6 is a halftone image displayed on a display showing a correlogram (a) and a line drawing correlogram (b) observed in the processing procedure of the present invention, and objects (c, d) obtained by reciprocally displaying them. .

FIG. 7 shows an example of a fetal heart rate diagram obtained by the method of the present invention.

Claims (3)

(57) [Claims] 1. A fetal monitoring apparatus for determining a fetal heart rate based on an autocorrelation of an envelope of an ultrasonic Doppler fetal heart rate signal, generates a correlogram of the filtered, detected, and filtered signal, and generates a correlogram of the correlogram. A fetal monitoring apparatus configured to determine the position of the first main maximum after performing the two-dimensional image processing above, and obtain a fetal heart rate value based on the position of the first main maximum. 2. The processing content of the two-dimensional image processing is as follows:
2. The device according to claim 1, characterized in that it is adaptively variable depending on the signal to be processed, the current and previous and past heart rate values, and the status of its change. 3. The method according to claim 1, wherein an extended correlogram is used as said correlogram.
The device described in 1.