EP1097439A1

EP1097439A1 - Method and device for detecting drifts, jumps and/or outliers of measurement values

Info

Publication number: EP1097439A1
Application number: EP99939929A
Authority: EP
Inventors: Martin Daumer
Original assignee: Individual
Current assignee: Daumer Martin Dr
Priority date: 1998-06-22
Filing date: 1999-06-22
Publication date: 2001-05-09
Anticipated expiration: 2019-06-22
Also published as: EP1097439B1; ATE261164T1; US6556957B1; WO1999067758A1

Abstract

A method for detection an alarm state of measurement signal values received by means for detecting measurement values, wherein an alarm state is triggered when for a currently received measurement signal value at least one pre-set limit value is exceeded, has a faster recognition of an alarm situation and at the same time a lower rate of false alarms when in a first step for measurement signal values which are subsequent in time their position parameter (2) and a corresponding deviation parameter (3) of the measurement signal values from the position parameter is calculated in an adjustable time window, wherein in a second step each further subsequent measurement signal value is compared to the position parameter (2) an weighted with the deviation parameter (3) in order to obtain a respective evaluation quantity, and wherein in a third step an outlier state (6) is detected when the evaluation quantity exceeds an adjustable outlier parameter, whereas an alarm state (8) indicating the presence of a significant drift or jump of the measurement signal values is detected when the evaluation quantity exceeds an adjustable alarm parameter.

Description

Method and device for detecting drifts, jumps and / or outliers of measured values

The invention relates to a method for the detection of an alarm state or for the detection of drifts, jumps and / or outliers of measurement signal values received via measurement data acquisition means, an alarm state being triggered when a currently received measurement signal value or a value derived from measurement values predetermined limit value or predetermined interval limits is exceeded. Examples of the extremely numerous applications of the method according to the invention are, in particular in the field of medicine, perioperative monitoring, monitoring vital parameters in intensive care units, sleep monitoring, CTG (cardio-tocography), and in other areas fire and smoke warning systems, acoustic monitoring systems, such as B. Baby monitor called.

Alarm systems in intensive care monitors, which typically display and analyze cardiovascular parameters (ECG, blood pressure), oxygen saturation (SpO2), gas exchange and metabolic parameters as well as EEG and EMG online, are intended to draw the attention of the attending physician or nurse to potentially life-threatening conditions for the direct monitored patients. An ideal alarm system would be characterized by the following properties, all of which are not optimally implemented in the prior art:

1. Low false alarm rate to avoid the undesirable effect of getting used to the alarm situation and to counteract the tendency to deactivate the alarm, which is often perceived as disturbing.

2. Short delays between the initiation of a critical situation and the triggering of the alarm in order to ensure that a time advantage for therapeutic interventions can be vital.

3. A high degree of adaptability, in order to avoid that too many parameters have to be preset by hand and readjusted during the treatment, and thus distract from the actual monitoring task. In particular, it should be possible to recognize several alarm situations which follow one another at a certain time interval. 4. A high degree of meaningfulness of the adjustable parameters to ensure that the alarm system can also be operated easily and without errors.

5. The greatest possible simplicity and thus the greatest possible computing speed, in order to avoid complex calculations which could only be carried out with expensive processors and memory elements, and to avoid any computing time restrictions.

6. Great informative value to enable differentiated reactions.

7. Integrated detection of outliers to enable a differentiation between life-threatening conditions, device and supply line failure and incorrect measurements.

8. Clear decision rules to ensure exportability and enable retrospective analysis and parameter correction.

Current alarm systems in intensive care medicine have a false alarm rate of 70% to 99.5%, depending on the monitored physiological

Size. The high false alarm rate leads to desensitization of the surveillance personnel and frequent manual alarm deactivation. The known alarm systems are triggered when the size to be monitored exceeds the preset upper or lower limits. Such alarm systems are referred to as threshold alarm systems. In order to lower the false alarm rate, the upper limit must be chosen rather high and the lower limit rather low, which inevitably leads to longer time delays in alarm-worthy situations. In addition, such an all-or-nothing system does not comply with the ISO standard, which proposes a graduated alarm system with different warning divisions.

In the known threshold alarm system, an upper and a lower threshold are specified for a fluctuating signal, an alarm being triggered when the signal moves from the interval defined by the threshold values. The threshold alarm has the following disadvantages. He is unstable towards

Runaway. It is not adaptive, ie limits have to be set manually and, in particular in the case of a signal with a drift, e.g. B. caused by a time be permanently adjusted before changing the detector sensitivity. If the threshold alarm limits are set too far, there will be long delays before an alarm is detected. If the limits are too narrow, false alarms often occur. In practice, therefore, a so-called "border balancing act" or an option, such as "all alarms off for two minutes", is set. Furthermore, the threshold alarm system is not suitable in the event that a large number of signals must be monitored by an alarm system.

With regard to the prior art, reference is made to the patent specification DE 35 23 232 C2.

From this publication, a fire alarm system for determining and delivering an analog value corresponding to a change in a physical appearance of the ambient conditions is known. There are a sampling device for sampling an analog detection signal emitted from a determination section within a certain period of time, a data processing device for forming an average value from the sampling data, and a storage device in which this sampling data can be stored and an alarm device which detects the presence of a fire after evaluation of the Indicates mean, provided. Characteristically, the data processing device is designed such that the scan data are written sequentially into the memory device, and a moving average is continuously formed from a certain number of the last stored scan data, the oldest scan data storage value in the sequence being replaced by the latest one .

Furthermore, DE 31 27 324 A1 discloses a method and an arrangement for increasing the sensitivity and interference immunity in a hazard, in particular fire alarm system.

In particular, it is not known from either of the above-mentioned publications,

To calculate scattering parameters from the successive measured values, so that the method used to trigger the alarm is adaptive and thus capable of learning. Therefore, the above methods are unable to adapt, for example, to a change in the sensitivity of the detector over time.

Finally, DE 44 17 574 C2 relates to patient alarm detection using a target mode. In this process, an intentional change a physiological parameter of a patient dynamic

Limits are defined and an alarm is generated if the measured parameter values are outside the dynamic limits. This document therefore only discloses a variant of the threshold alarm known per se.

The object of the present invention is therefore to avoid the disadvantages of the prior art and, in particular, to further develop a method of the type mentioned at the outset in such a way that an "alarm situation" is recognized more quickly and with a lower false alarm rate compared to the prior art.

In technical terms, the object is achieved in that, in a first step, the mean value and the corresponding scatter of these measurement signal values are calculated from the mean value in a settable time window for temporally successive measurement signal values, that in a second step each further subsequent measurement signal value for obtaining a the respective evaluation variable is compared with the mean value and weighted with the scatter, and in a third step an outlier state is detected in the case of an evaluation variable exceeding an adjustable outlier parameter, while in an evaluation quantity exceeding an adjustable outlier parameter

Evaluation variable an alarm state indicating the presence of a significant drift or jump in the measurement signal values is detected.

A distinction is thus made between two phases in the method according to the invention, a time window being provided in a first phase in which the characteristic course of the measurement signal values recorded therein is evaluated, the statistical average and the fluctuation range of the recorded measurement signal values being determined by this average value . In the second phase of the method according to the invention, the currently received measurement signal values are compared with the mean value and the scatter representing the fluctuation range, the evaluation variable determined thereby representing a measure of the presence of a significant drift. By including the temporal development of the measurement signal values recorded in the time window in this evaluation variable thus obtained, there is an overall higher degree of reliability in the detection of alarm states than methods according to the prior art, so that a lower false alarm rate can be achieved. This is based in particular on the automatic tracking of the interval limits. Due to the Because of the method according to the invention, a distinction between outliers and outliers leading to falsifications and / or false alarms and alarm states is made possible in intensive care applications, on the one hand, to differentiate between life-threatening conditions and, on the other hand, equipment or supply line failures leading to incorrect measurements, which leads to a further reduction in the false alarm rate .

An advantage of the method according to the invention is that on-line detection of outliers is provided. It is also advantageous that the method according to the invention is adaptive, i. H. for example only physiological

Limits have to be preset. Furthermore, according to the invention, drifts and / or jumps can be automatically recognized. Finally, the method according to the invention only has a short delay time.

In order to achieve a high computing speed, the evaluation variable is determined by forming the difference between the measurement signal value and the calculated mean value and then normalizing the difference. The evaluation variable is weighted by forming a division from the normalized difference between the measurement signal value and the mean value with the calculated scatter.

According to an advantageous embodiment of the method according to the invention, an outlier state is detected when the normalized difference between the measurement signal value and the mean value, weighted with the calculated scattering, exceeds the set outlier parameter. On the other hand, an alarm state is detected if the normalized difference between the measured signal value and the mean value, weighted with the calculated scatter, exceeds the set alarm parameter.

In order to eliminate measurement errors that occur, for example, due to device failure or measurement artifacts, the corresponding measurement signal value is replaced by the current mean value calculated in the time-shifted window when an outlier condition occurs and the next following measurement signal value is processed.

Alternatively, a different type of replacement can be carried out, which is particularly preferred for statistical reasons. It can for example, adding a noise or performing another imputation. The outlier value can in particular be replaced by an average plus an added random number, which originates from a probability distribution. Finally, such a falsified or falsified measurement value can also simply be ignored for the further calculation.

It has proven to be expedient if the mean value of the successive measurement signal values is formed from the summation of the individual measurement signal values, the number of the summation steps being determined by the width of the time window. The standard deviation is used as the scatter, the number of summation steps being determined by the width of the time window.

A further development of the method according to the invention, which is particularly advantageous from a computational point of view, is that the time window is positioned by means of a time delay in order to be able to recognize even small gradients in the course of the measured variable over time, so that long-term drifts due to a correspondingly distant one , delayed moving window. Short-term drifts can also be detected with a correspondingly close, delayed moving window.

To make it easier to distinguish between outlier states and alarm states, the outlier parameter is set to a higher value than the alarm parameter.

It has proven to be particularly expedient if the width of the time window is preferably set to 10 successive measurement signal values and the outlier parameter is set to 6 and the alarm parameter to 3.

In terms of device technology, the above-mentioned object is achieved in a device with a measured value detection device for receiving measured value signals and a measured value transmission device for converting and processing the received measured value signals, and one when a measured value is exceeded

Alarm device which can be triggered by the limit value is solved in that, in order to detect the measurement signal values, in a time window which can be set according to width and time delay. A memory device is provided that in an initialization phase for temporally successive measurement signal values in the adjustable time window, calculation means are provided for calculating the mean values and the corresponding scatter, and that in a process phase a processor device is provided for obtaining an evaluation variable, with an adjustable alarm parameter exceeding evaluation size actuated the alarm device.

By means of the interaction of the individual components, outlier states and alarm states can thus be distinguished from one another in accordance with an evaluation variable thereby obtained, so that the false alarm rate can thus be significantly reduced compared to methods according to the prior art.

An embodiment of the present invention will be explained below with reference to the accompanying drawings. In partially schematic views:

1 shows a flow chart with the essential process steps of the method according to the invention;

2 shows a measured value spectrum of the temporal development of a physiological measured variable;

3a shows a highly schematic illustration of a drift;

3b is a highly schematic representation of a jump; and

3c shows a highly schematic illustration of an outlier.

The method according to the invention, which is preferably implemented as a software program, is illustrated with its essential process steps in a flow diagram, designated as a whole by 10, in FIG. 1. During an initialization phase 1, a time window is provided in which an average value 2 and the associated scatter 3 of the measurement signal values around this average value are calculated over a length of i successive steps for the measurement signal values recorded in the time window. However, the mean value is not calculated from a series of the immediately preceding measured values, but that from a time window of latitude ω in the past with the selectable time delay d. The lower summation limit for determining the mean value thus results from the subtraction nd-ω, where n denotes the number of time steps carried out, d the time delay and ω the window width. In contrast, the upper summation limit results from the subtraction nd, so that the summation index i runs from nd-ω to nd. The same summation limits apply to the determination of the scatter 3.

In the actual process phase, an incrementation is carried out in a process step 4. In a further process step 5, the measurement signal value Y _{n acquired} in a specific time step is compared with the mean value determined in the initialization phase by carrying out a difference formation and providing this difference formation with an amount normalization. In order to also take into account the variance in the comparison, the amount-standardized difference is weighted with the variance by including the variance as a divisor. The evaluation variable thereby obtained serves as a measure for the detection of outlier states occurring in this process step 4.If the evaluation variable obtained for the currently recorded measurement signal value is greater than a preset outlier parameter o (o> 0), the query in process step 4 reveals that an outlier state 6 is present. The outlier state can be ignored for the following calculation or replaced by a "reasonable" value. Imputation procedures are particularly suitable for this. In this case, the sequence program returns to increment instruction 4.

If the query in block 5 yields a negative result, it is determined in query block 7 whether the evaluation variable obtained for the currently recorded measurement signal value is greater than a preset alarm parameter a. If the result is positive, an alarm state 8 is present. In the exemplary embodiment, a return to the initialization phase is carried out, while if the result is negative, a return is made to the increment instruction. As a boundary condition when distinguishing between outlier states and alarm states, the outlier parameter is assigned a higher value than the alarm parameter. By differentiating between outlier states and alarm states, a differentiation between significant states and erroneous measurements is achieved, where erroneous measurements can result from supply failures or measurement artifacts. The detection and elimination of such incorrect measurements thus leads to a reduction in false alarms.

2 shows the course over time of a physiological measured variable. The abscissa axis serves as the time axis τ _C p, while the ordinate axis represents the amplitude of the measurement signal.

3a shows a highly schematic illustration of a drift. 3b shows a highly schematic representation of a jump. 3c shows a highly schematic representation of an outlier. The time dependence of a measured signal is shown.

In summary, the following is characteristic of the method according to the invention: The internal parameters of the algorithm are the window width ω (ω> 0), the delay d (d> 0), the initialization length i (i> ω + d), the outlier parameter o (o> 0) and the alarm parameter a (a> 0).

After an initialization phase of the length of i time steps, the newly measured value is compared with an average value estimated from the previous measured values together with the associated scatter (the empirical standard deviation) - in this respect, the algorithm is a natural generalization of the normal threshold alarm, in which the mean value and the spread are known be assumed. However, the mean value is not calculated from a series of the immediately preceding measured values, but from a time window of width ω in the past, with the selectable time delay d. This

The type of calculation circumvents the problem that the measured values used to estimate the mean value and the spread have already started to drift and thus contribute to a considerable bias, which can go so far that a sufficiently slow drift is not recognized at all. Rather, the freely selectable delay d gives the option of choosing the critical angle of the slope that is just about to be recognized. Naturally, the smaller the slope, the larger d must be chosen. Each newly measured value is compared with the current mean value estimated according to the method according to the invention as follows: the measured value is more than the product of the selectable outlier factor and the scatter of the estimated mean value removed, it is classified as an outlier and replaced for further calculations by the current mean value (plus a random number with an expected value of zero and scatter according to the estimated scatter). If this is not the case, but the measured value is more than the product of (selectable) alarm factor a and scatter from the estimated mean, it is output that there is a significant drift, depending on the direction of the deviation, a drift upwards or below. In all other cases, no message is issued. Then the next time step is processed. You can choose whether you want to reinitialize after an alarm is issued, possibly with a further selectable time delay, or whether you want to continue calculating without a new initialization. The window width ω influences the fluctuations of the estimated mean - the fluctuations decrease proportionally to the root of ω.

For many purposes, the following values prove to be favorable starting values that can then be optimized: window width ω> 10, outlier parameter o = 6 and the alarm parameter a = 3. The calculated information outliers yes / no, alarm for drift up / down, or no significant drift can either be output directly on the screen or acoustically via agreed sound sequences, or at the entrance to an intelligent alarm system.

The invention was previously explained in more detail using a preferred exemplary embodiment. However, it is obvious to a person skilled in the art that various modifications and modifications can be made without departing from the concept on which the invention is based. In particular, it should be noted that in the present description under the expression "position parameters" in particular an average, median, or the like. and under the expression "scattering parameters" a standard deviation, quantile, or the like. is understood.

Claims

claims

1. A method for recognizing an alarm state of measurement signal values received via measurement value acquisition means, an alarm state being triggered when at least one predetermined limit value for a currently received measurement signal value is exceeded, characterized in that, in a first step, their position parameters for temporally successive measurement signal values in an adjustable time window (2) and a corresponding scatter parameter (3) of these measurement signal values is calculated by the position parameter, that in a second step each further subsequent measurement signal value is compared with the position parameter (2) to obtain a respective evaluation variable and weighted with the scatter parameter (3) , and that in a third step an outlier condition (6) is detected in the case of an evaluation parameter exceeding an adjustable outlier parameter, whereas in the case of an evaluation parameter exceeding an adjustable alarm parameter, the prelim an alarm state (8) indicating a significant drift or jump in the measurement signal values is detected.

2. The method according to claim 1, characterized in that the evaluation variable is determined by forming the difference between the measurement signal value and the calculated position parameter (2) with subsequent normalization of the difference.

3. The method according to claim 2, characterized in that the

The evaluation variable is weighted by forming a division from the standardized difference between the measurement signal value and the position parameter (2) with the calculated scattering parameter (3).

4. The method according to claim 3, characterized in that a

Outlier condition (6) is detected when the normalized difference between the measured signal value and the position parameter (2) weighted with the calculated scattering parameter (3) exceeds the set outlier parameter.

5. The method according to claim 3 or 4, characterized in that an alarm state (8) is detected when the calculated with the

Scattered parameter (3) weighted normalized difference between measurement signal value and position parameter (2) exceeds the set alarm parameter.

6. The method according to any one of claims 1 to 5, characterized in that when an outlier condition (6) occurs, the corresponding measurement signal value is replaced by a replacement value and the next measurement signal value is processed.

7. The method according to claim 6, characterized in that the replacement value is the current position parameter (2).

8. The method according to any one of claims 1 to 5, characterized in that when an outlier condition (6) occurs, the corresponding measurement signal value is omitted for the further calculation and the next measurement signal value is processed.

9. The method according to any one of claims 1 to 8, characterized in that the position parameter of the successive measurement signal values is formed from the summation of the individual measurement signal values, the number of summation steps being determined by the width of the time window.

10. The method according to any one of claims 1 to 9, characterized in that the standard deviation is used as the scattering parameter (3), the number of summation steps being determined by the width of the time window.

11. The method according to any one of claims 1 to 9, characterized in that a variable positioning of the time window is carried out by means of a time delay.

12. The method according to any one of claims 1 to 11, characterized in that the outlier parameter is set to a higher value than the alarm parameter.

13. The method according to any one of claims 9 to 12, characterized shows that the width of the time window is preferably set to ten consecutive measurement signal values and the outlier parameter is set to six and the alarm parameter to three.

14. Device for performing the method according to one of claims 1 to 13.

15. The apparatus of claim 14, with a measured value detection device for receiving measured value signals and a measured value transmission device for converting and processing the received

Measured value signals and an alarm device which can be triggered when at least one limit value is exceeded, characterized in that a memory device is provided for detecting the measured signal values in a time window which can be set according to width and time delay, and that calculation means for calculating the position parameters (2 ) and the corresponding scatter (3) are provided, and that a processor device is provided for obtaining an evaluation variable which actuates the alarm device when an evaluation variable exceeds an adjustable alarm parameter.