Process and signal processing unit for determining a pneumatic parameter with the use of a lung-mechanical model and of a gradient model

Abstract

A process and a signal processing unit determine a pneumatic parameter (P.sub.mus) for the spontaneous breathing of a patient. The patient is ventilated mechanically by a ventilator. A lung-mechanical model (20) and a gradient model (22) are preset. The lung-mechanical model (20) describes a relationship between the pneumatic parameter (P.sub.mus) as well as a volume flow signal (Vol), a volume signal (Vol) and/or a respiratory signal (Sig), which can be measured. The gradient model (22) describes a value for the pneumatic parameter (P.sub.mus) as a function of N chronologically earlier values of the pneumatic parameter (P.sub.mus) or of a variable correlating with the pneumatic parameter (P.sub.mus). N values for the correlating variable are determined at first. At least one additional value is subsequently determined for the pneumatic parameter (P.sub.mus). N chronologically earlier values of the correlating variable, current signal values, the lung-mechanical model (20) and the gradient model (22) are used for this purpose.

Claims

1. A process comprising the steps of: presetting a computer-analyzable lung-mechanical model which describes at least one relationship between a pneumatic parameter to be determined and one or more of: a volume flow signal for a flow of breathing air to and/or from the lungs of the patient; and a volume signal for a filling level of the lungs of the patient; and a measurable pneumatic pressure signal for pressure in airways or in an esophagus of the patient; and a respiratory signal correlated with the spontaneous breathing of the patient; presetting a computer-evaluable gradient model for the pneumatic parameter, wherein the gradient model describes a value of the pneumatic parameter for a scanning time as a function of at least N values of a variable, which is variable over time and which is correlated with the pneumatic parameter for N previous scanning times, wherein N is a preset number, and automatically generating, with a data-processing signal processing unit, a respective signal value for at least one signal to which the lung-mechanical model refers; and during the generation of the signal values, with the signal processing unit, receiving measured values from an airway pressure sensor and/or from an esophageal pressure sensor and/or from a volume flow sensor and/or from a volume sensor and based on the measured values generating one or more of the volume flow signal and the volume signal and the measurable pneumatic pressure signal; during the generation of the signal values, with the signal processing unit, receiving sensor array measured values from a sensor array on skin of the patient and generating the respiratory signal from the sensor array measured values; with the signal processing unit, determining N respective values for any variable that is correlated with the pneumatic parameter and appears in the gradient model for N consecutive scanning times; and with the signal processing unit, determining a respective value for the pneumatic parameter during a subsequent use phase for at least one scanning time, wherein the signal processing unit uses for the determination of the respective value for the pneumatic parameter during the subsequent use phase at least: generated signal values for the scanning time; respective, already determined N values of the correlating variable of the gradient model for N previous scanning times; the lung-mechanical model; and the gradient model, wherein the signal processing unit provides the pneumatic parameter as input to a ventilator for synchronizing operation of the ventilator with spontaneous breathing of a ventilated patient based on the pneumatic parameter.

2. A process in accordance with claim 1, wherein: during the subsequent use phase, in the step of determining a value of the pneumatic parameter for at least one scanning time, the signal processing unit uses at least one preset computer-evaluable Kalman filter, which comprises the gradient model; and a sequence of N values of the correlating variable or of each correlating variable of the gradient model is used as a component of a state vector of the Kalman filter for N consecutive scanning times.

3. A process in accordance with claim 1, wherein: during the subsequent use phase, in the step of determining a value of the pneumatic parameter for a scanning time, the signal processing unit uses a preset computer-analyzable second Kalman filter, which comprises the lung-mechanical model; and values for the pneumatic parameter and values for one or more of the volume flow signal and the volume signal and the respiratory signal are used as components of an observation equation of the Kalman filter.

4. A process in accordance with claim 1, wherein: during the subsequent use phase, in the step of determining a value of the pneumatic parameter for a scanning time, the signal processing unit uses at least one preset computer-analyzable Kalman filter, which comprises both the gradient model and the lung-mechanical model; a sequence of N values of the correlating variable or of each correlating variable of the gradient model for N consecutive scanning times and values for at least one of the volume flow signal and of the volume signal and of the pneumatic pressure signal and of the respiratory signal are used as components of an observation equation of the Kalman filter.

5. A process in accordance with claim 1, wherein the preset lung-mechanical model comprises: a first relationship between the pneumatic parameter as well as one or more of the volume flow signal and the volume signal and the pneumatic pressure signal; and a second relationship between the pneumatic parameter and the respiratory signal, wherein the signal processing unit: determines a respective first value and a second value for the pneumatic parameter for N scanning times during the subsequent use phase; uses the first relationship as well as at least one respective value of the volume flow signal and of the volume signal and of the measurable pneumatic pressure signal for the determination of the first value for the scanning time; and uses the second relationship as well as at least one value of the respiratory signal for the determination of the second value for the scanning time; and uses, during the subsequent use phase in the step of determining a value of the pneumatic parameter for a scanning time: the N first values; the N second values; and the gradient model.

6. A process in accordance with claim 1, wherein when no value is available for a signal occurring in the lung-mechanical model at a scanning time, the value of the pneumatic parameter is determined for said scanning time exclusively with the use of the gradient model.

7. A process in accordance with claim 1, wherein the gradient model describes the value of the pneumatic parameter at the scanning time as a function of N values of the correlating variable for N previous scanning times as well as additionally of the respective value of at least one measurable variable at the scanning time.

8. A process in accordance with claim 1, wherein the preset lung-mechanical model comprises at least one model parameter, which is variable over time, wherein during the subsequent use phase, in the step of determining a value of the pneumatic parameter for a scanning time, the signal processing unit calculates a respective model parameter value with the use of signal values for said scanning time and of signal values for at least one earlier scanning time for at least one model parameter, which is variable over time, and additionally determines at least one value for the pneumatic parameter with the use of the calculated model parameter value of each calculated model parameter value.

9. A process in accordance with claim 1, wherein during an initialization phase, the signal processing unit determines N values of the variable correlating with the pneumatic parameter for N consecutive scanning times with the use of generated signal values.

10. A process in accordance with claim 1, wherein the signals comprise one or more of an EMG signal acquired by one or more sensors on skin of the patient, an MMG signal acquired by one or more sensors on skin of the patient, and a pneumatic signal generated by a sensor in the body of the patient.

11. A process in accordance with claim 1, further comprising providing a computer program, which is executed on the signal processing unit, upon the signal processing unit receiving measured values from at least one of the airway pressure sensor, the esophageal pressure sensor, the volume flow sensor, the volume sensor and the sensor array, wherein an execution of the computer program on the signal processing unit causes the signal processing unit to carry out at least some of the process steps.

12. A process in accordance with claim 1, wherein a signal sequence, comprising commands to be executed on the signal processing unit causes the signal processing unit to carry out at least some of the process steps upon the signal processing unit receiving measured values from at least one of the sensor array for measuring the spontaneous breathing of the patient, and from the airway pressure sensor and from the esophageal pressure sensor and from the volume sensor.

13. A signal processing unit for an approximate automatic determination by calculation of a pneumatic parameter for spontaneous breathing of a patient, the signal processing unit comprising: a processor comprising at least occasional read access to a memory with a stored computer-analyzable lung-mechanical model and a stored computer-analyzable gradient model for the pneumatic parameter, wherein: the lung-mechanical model describes at least one relationship between the pneumatic parameter and one or more of a volume flow signal for a flow of breathing air to and/or from the lungs of the patient, and a volume signal for a filling level of the lungs of the patient and a measurable pneumatic pressure signal for pressure in the airways or in an esophagus of the patient and a respiratory signal which is correlated with the spontaneous breathing of the patient; the gradient model describes a value of the pneumatic parameter for a scanning time at least as a function of N values of a variable, which is variable over time and is correlated with the pneumatic parameter for N previous scanning times, wherein N is a preset number; the processor is configured to generate a respective signal value repeatedly for at least one signal to which the lung-mechanical model refers, the processor is configured to receive measured values from at least one of an airway pressure sensor and an esophageal pressure sensor and a volume flow sensor and a volume sensor to generate from the received measured values the volume flow signal and the volume signal and the pressure signal and measured values from a sensor array which measures a variable that is correlated with the spontaneous breathing of the patient, and a second sensor array on the skin of the patient and from a sensor in the body of the patient when generating the signal values; the processor is configured to determine, during an initialization phase, N values of the variable or each variable correlating with the pneumatic parameter and occurring in the gradient model for N different scanning times; the processor is further configured to determine in a subsequent use phase a respective value for the pneumatic parameter for at least one scanning time, the processor is configured to use for this determination during the subsequent use phase at least: generated signal values for the at least one scanning time; respective, already determined N values of the variable or each variable of the gradient model, which variable is correlated with the pneumatic parameter, for N previous scanning times, the lung-mechanical model; and the gradient model; the processor is configured to provide the pneumatic parameter as output to a ventilator for synchronizing operation of the ventilator with the spontaneous breathing of the patient.

14. A system comprising the signal processing unit in accordance with claim 13, and further comprising the ventilator, wherein the ventilator is configured: to ventilate a patient mechanically; and to use the pneumatic parameter for the spontaneous breathing of the patient for the mechanical ventilation, wherein the pneumatic parameter used has been determined by the signal processing unit.

15. A system in accordance with claim 14, wherein: during the subsequent use phase, the signal processing unit uses at least one preset Kalman filter, which comprises the gradient model, for the determining of a value of the pneumatic parameter for at least one scanning time; and a sequence of N values of the correlating variable or of each correlating variable of the gradient model is used as a component of a state vector of the Kalman filter for N consecutive scanning times.

16. A system in accordance with claim 14, wherein: during the subsequent use phase, for determining a value of the pneumatic parameter for a scanning time, the signal processing unit uses a preset second Kalman filter, which comprises the lung-mechanical model; and values for the pneumatic parameter and values for one or more of the volume flow signal and the volume signal and the respiratory signal are used as components of an observation equation of the Kalman filter.

17. A system in accordance with claim 14, wherein: during the subsequent use phase, for determining a value of the pneumatic parameter for a scanning time, the signal processing unit uses at least one preset Kalman filter, which comprises both the gradient model and the lung-mechanical model; a sequence of N values of the correlating variable or of each correlating variable of the gradient model for N consecutive scanning times and values for at least one of the volume flow signal and of the volume signal and of the pneumatic pressure signal and of the respiratory signal are used as components of an observation equation of the Kalman filter.

18. A system in accordance with claim 14, wherein the lung-mechanical model comprises: a first relationship between the pneumatic parameter as well as one or more of the volume flow signal and the volume signal and the pneumatic pressure signal; and a second relationship between the pneumatic parameter and the respiratory signal, wherein the signal processing unit: determines a respective first value and a second value for the pneumatic parameter for N scanning times during the subsequent use phase; uses the first relationship as well as at least one respective value of the volume flow signal and of the volume signal and of the measurable pneumatic pressure signal for the determination of the first value for the scanning time; and uses the second relationship as well as at least one value of the respiratory signal for the determination of the second value for the scanning time; and uses, during the subsequent use phase in the step of determining a value of the pneumatic parameter for a scanning time: the N first values; the N second values; and the gradient model.

19. A system in accordance with claim 14, wherein the preset lung-mechanical model comprises at least one model parameter, which is variable over time, wherein during the subsequent use phase, in the step of determining a value of the pneumatic parameter for a scanning time, the signal processing unit calculates a respective model parameter value with the use of signal values for said scanning time and of signal values for at least one earlier scanning time for at least one model parameter, which is variable over time, and determines at least one value for the pneumatic parameter with the use of the calculated model parameter value.

20. A system in accordance with claim 14, wherein during the initialization phase, the signal processing unit determines N values of the variable correlating with the pneumatic parameter for N consecutive scanning times with the use of generated signal values.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) In the drawings:

(2) FIG. 1 is a schematic view showing which sensors measure which different variables for determining the pneumatic parameter P.sub.mus;

(3) FIG. 2 is a schematic view showing how a parameter estimator interacts automatically with a P.sub.mus estimator;

(4) FIG. 3 is a schematic view showing an additional estimator, which automatically combines two estimations determined in different ways for the pneumatic parameter P.sub.mus; and

(5) FIG. 4 is a graph showing an example of how the process noise and the measurement noise are taken into account.

DESCRIPTION OF PREFERRED EMBODIMENTS

(6) Referring to the drawings, the present invention is used in the exemplary embodiment to determine a pneumatic parameter P.sub.mus, which is correlated with the spontaneous breathing (intrinsic breathing activity) of a patient P. This pneumatic parameter P.sub.mus is variable over time, i.e., P.sub.mus=P.sub.mus(t). This patient P is ventilated mechanically by means of a ventilator 1 at least from time to time, as a result of which a fluid connection is established from the ventilator 1 to the patient P, and this connection is a closed ventilation circuit between the ventilator 1 and the patient P in one embodiment. While the patient P is being ventilated mechanically, the mechanical ventilation, which is brought about by the ventilator, is superimposed to his spontaneous breathing, or the patient is fully sedated from time to time, i.e., he has no spontaneous breathing.

(7) The knowledge of the pneumatic parameter P.sub.mus is preferably used to regulate the ventilator 1, for example, in order for the ventilator 1 to carry out ventilation strokes synchronously with the spontaneous breathing of the patient P. It is also possible to output the determined pneumatic parameter P.sub.mus continuously in a form perceptible by a human being and/or to generate and output alarms as a function of the pneumatic parameter P.sub.mus. It is possible that the pneumatic parameter P.sub.mus determined is used both to synchronize the ventilation strokes of the ventilator 1 with the spontaneous breathing of the patient P and also to output the pneumatic parameter P.sub.mus in a form perceptible by a human being

(8) The pneumatic parameter P.sub.mus cannot be measured directly especially in case of mechanical ventilation. The pneumatic parameter P.sub.mus, which is variable over time, is rather determined according to the present invention indirectly, and the values of a plurality of variables, which occur in the fluid connection from the ventilator 1 to the patient P, are used for the determination. These variables are called signals and the values of the variables are called signal values. A signal shall hereinafter be defined as the course in the time range or even in the frequency range of a directly or indirectly measurable variable, which is variable over time, and which is correlated with a physical variable. This physical variable is connected here with the spontaneous breathing and/or with the mechanical ventilation and/or with the cardiac activity of the patient P and is generated by at least one signal source in the body of the patient P and/or by the ventilator 1. A respiratory signal is correlated with the spontaneous breathing and/or with the mechanical ventilation of the patient P, and a cardiogenic signal is correlated with the cardiac activity of the patient P.

(9) The measured values, which originate from sensors at or in the patient P or close to the patient P or in the fluid connection from the ventilator 1 to the patient P, are processed at each scanning time t.sub.i. A set of signal values, which set comprises the values of different signals at this scanning time is generated thereby. Using at least one set of signal values, preferably a plurality of sets of signal values, the signal processing unit 5 determines an estimated value P.sub.mus,est(t.sub.i) for the pneumatic parameter P.sub.mus. The subscript est indicates that the determined signal values and the value for P.sub.mus are estimated and do, as a rule, deviate from the actual values.

(10) FIG. 1 shows schematically which signals can be generated from measured values, for which the measured values are processed in a suitable manner. Shown are: the patient P ventilated mechanically at least from time to time, the esophagus Sp, the stomach Ma and the diaphragm Zw of the patient P, a flexible connection piece 3, which is located in the mouth of the patient P during the ventilation, a flexible measuring catheter 8, which is introduced into the esophagus Sp of the patient and which begins in the connection piece 3, the ventilator 1, which ventilates the patient mechanically at least from time to time and comprises a data-processing signal processing unit 5, which has read access at least from time to time to a memory 9, a sensor array with two sets 2.1.1 through 2.2.2 of sensors with at least two respective measuring electrodes each, wherein the measuring electrode sets 2.1.1 and 2.1.2 are arranged close to the sternum of the patient P and the measuring electrode set 2.2.1 and 2.2.2 are arranged close to the costal arch, and wherein the sensor array comprises, in addition, at least one reference electrode, not shown, for grounding, a pneumatic sensor 7, which is located at a distance in space from the body of the patient P, an optional sensor 4, which comprises an image recording device and an image analysis unit and is directed towards the thoracic region of the patient P, an optional pneumatic sensor 6 in the form of a probe or of a balloon in the esophagus Sp and close to the diaphragm Zw of the patient P, wherein the sensor 6 measures a pressure P.sub.es (pressure in esophagus) in the esophagus Sp and is in fluid connection with the measuring catheter 8, and an optional gastric probe 14 in the stomach Ma of the patient P, which is likewise in a fluid connection with the measuring catheter 8.

(11) The pneumatic sensor 7 comprises a transducer 7.1 comprising an opening, which is arranged in the vicinity of the mouth of the patient P and taps air from the fluid connection. The tapped air is transmitted via a flexible tube (suggested by arrows) to a pressure sensor 7.2, which measures a parameter for the airway pressure P.sub.aw (pressure in airway) in the fluid connection and a parameter for the volume flow Vol. In one embodiment, the transducer 2.1 is arranged in or at a Y-piece close to the connection piece 3, i.e., close to the mouth of the patient P. It is also possible that the pneumatic sensor 7 comprises two individual sensors, namely, a sensor for the airway pressure P.sub.aw and a sensor for the volume flow Vol.

(12) The measuring electrode sets 2.1.1 through 2.2.2 of measuring electrodes as well as the reference electrode, not shown, yield after signal processing an electrical respiratory signal Sig, which is correlated with the pneumatic parameter P.sub.mus to be determined. Instead of an electrical respiratory signal (EMG signal), it is also possible to generate and use a signal in the form of a mechanomyogram (MMG signal). As a rule, the EMG signal or the MMG signal is generated by a superimposition of a respiratory signal and of a cardiogenic signal. The signal processing unit 5 compensates the influence of the cardiogenic signal by calculation to the extent possible, for example, as described in DE 10 2015 015 296 A1 (corresponding US 2018344194 (A1) is incorporated herein by reference), in DE 10 2007 062 214 B3 (corresponding U.S. Pat. No. 8,109,269 (B2) is incorporated herein by reference) or in M. Ungureanu and W. M. Wolf: Basic Aspects Concerning the Event-Synchronous Interference Canceller, IEEE Transactions on Biomedical Engineering, Vol. 53, No. 11 (2006), pp. 2240-2247 (which is incorporated herein by reference).

(13) The pneumatic sensor 7 yields as the signalafter signal processingthe airway pressure P.sub.aw (pressure in airway) in front of the mouth of the patient P. As long as the patient P is being ventilated mechanically, this airway pressure P.sub.aw results from a superimposition of the spontaneous breathing of the patient P and the mechanical ventilationor even from the mechanical ventilation only. The optional pneumatic sensor 6 yields as the signalafter signal processingthe pressure in the esophagus P.sub.es (pressure in esophagus), which likewise results from a superimposition of spontaneous breathing and mechanical ventilation. The optional gastric probe 14 measures the gastric pressure P.sub.ga in the stomach Ma of the patient P.

(14) Different signals may also be used as the respiratory signal Sig or as a respiratory signal Sig. It is also possible to use two respiratory signals Sig.sub.1 and Sig.sub.2. In one embodiment the electrical signal from the measuring electrodes 2.1.1 through 2.2.2 on the skin of the patient P is used as the respiratory signal Sig or as a respiratory signal Sig. In another embodiment, a signals from the pneumatic sensor 6 and/or from the gastric probe 14 are used as the respiratory signal Sig or as a respiratory signal Sig.

(15) In addition, the volume flow Vol of breathing air into and out of the lungs of the patient P and/or the volume Vol of the lungs of the patient P can be derived as signals from the measured values of the pneumatic sensor 7, 14, of the optional pneumatic sensor 6 and/or of the optional optical sensor 4. It is possible to calculate the lung volume Vol by numerical integration over the measured volume flow Vol. It is also possible to derive the lung volume Vol from measured values of the optical sensor 4, on the one hand, and to obtain it by numerical integration, on the other hand, and then to average in a suitable manner between the two signals determined in different ways.

(16) A preset, computer-analyzable lung-mechanical model 20 and a preset, computer-analyzable gradient model 22 are stored in the memory 9, to which the signal processing unit 5 has read access at least from time to time. The signal processing unit 5 is capable of applying both models automatically to signal values. The lung-mechanical model 20 has at least one and preferably more model parameters, which are, as a rule, variable over time and for which the signal processing unit 5 calculates at least once a respective value each, preferably one value for a plurality of scanning times, in order to take changes over time into account.

(17) FIG. 2 schematically shows the interaction of a parameter estimator 10 with a P.sub.mus estimator 11. Both the parameter estimator 10 and the P.sub.mus estimator 11 are implemented on the signal processing unit 5 and are preferably implemented by means of a software program. The parameter estimator 10 has read access to the preset lung-mechanical model 20 and optionally to computer-analyzable weightings 21 and it calculates values for the parameters of the lung-mechanical model 20, which are variable over time. A weighting estimator 12 has calculated these weightings 21, which will be described below. The weightings 21 may be variable over time. The P.sub.mus estimator 12 has read access to the preset gradient model 22 for P.sub.mus and receives the model parameter values, which the parameter estimator 11 has calculated. The P.sub.mus estimator 12 calculates estimated values P.sub.mus,est (t.sub.i) for the pneumatic parameter P.sub.mus to be determined, which refer to scanning times L.

(18) The signal processing unit 5 uses these values P.sub.mus,est(t.sub.i) in one embodiment in order to actuate the ventilator 1, e.g., for a proportional control, i.e., the intensity of the mechanical ventilation, especially the pressure in the fluid connection to the patient P, which pressure is brought about by the ventilation strokes and is variable over time, is proportional to the determined current spontaneous breathing of the patient P and is preferably proportional to P.sub.mus. In one embodiment, this estimated pneumatic parameter P.sub.mus is fed back again to the parameter estimator 10 in order to improve future estimations of P.sub.mus.

(19) Consequently, in the exemplary embodiment, the lung-mechanical model 20 is used, on the one hand, which comprises at least one relationship, preferably in the form of a model equation, between the pneumatic parameter P.sub.mus sought and a plurality of measurable signals, wherein the relationship or each relationship connects different signal values for the same scanning time. A measurable signal is defined as a variable, which is, as a rule, variable over time and which can be derived by signal processing from measured values of sensors, in this case, for example, from the sensors 4, 6, 7 and/or 14.

(20) This lung-mechanical model 20 contains in the exemplary embodiment at least one model parameter and preferably more model parameters, wherein the model parameter or each model parameter is, as a rule, likewise variable over time. The lung-mechanical model is preferably linear in each model parameter, but it is not necessarily linear in the signals. The model parameters do, however, change their values more slowly than does the pneumatic parameter P.sub.mus to be determined. This lung-mechanical model 20 has, of course, only an approximate validity.

(21) In one embodiment, the lung-mechanical model 20 comprises the model equation
P.sub.aw(t)=R*Vol(t)+E*Vol(t)+P.sub.mus(t)+P0+1.sub.n.(1)

(22) This model equation (1) has three model parameters, namely, the two lung-mechanical factors R and E as well as the summand P0. The factor R (resistance) describes the breathing resistance, which the airways of the patient P offer against the volume flow of the air flowing into and out of the lungs. The factor E (elastance) describes the elasticity of the lungs. The summand P0 (pulmonary end-expiration pressure, PEEP) describes the effect of an incomplete exhalation by the patient P. The summand 1.sub.n results from the inevitable process noise (deviation between model and reality) and is preferably treated as a normally distributed random variable with the expected value zero. The estimated values for the model parameters are not necessarily good estimations for the values of the physical lung parameters, but they are used above all to estimate the pneumatic parameter P.sub.mus.

(23) It is also possible to use the following model equation:
P.sub.aw(t)=R.sub.1*Vol(t)+R.sub.2*Vol(t)*|Vol(t)|+E*Vol(t)+P.sub.mus(t)+P0+1.sub.n.(2)
|x| designates the absolute value of x.
This equation can be rewritten as follows:
P.sub.aw(t)=R(Vol)*Vol(t)+E*Vol(t)+P.sub.mus(t)+P0+1.sub.n,(3)
in which R(Vol)=R.sub.1+R.sub.2*|Vol(t)|.

(24) The model parameters can be estimated more rapidly in some cases with equation (3) than with equation (2).

(25) In a variant, all model parameters R, E and P0 in the model equation (1) are assumed to be approximately constant over time. The model equation (1) is differentiated once according to time, as a result of which the summand P0 assumed to be constant disappears. This procedure leads to the following lung-mechanical model equation:
P.sub.aw(t)=R*Vol(t)+E*Vol(t)+P.sub.mus+1.sub.n.(4)

(26) The corresponding procedure can be applied to the model equations (2) and (3).

(27) It is also possible that the pressure P.sub.es in the esophagus Sp of the patient P is measured, doing so by means of the probe 6. This embodiment can replace the embodiment in which the airway pressure P.sub.aw is measured by means of the sensor 7. It is also possible that the sensor 7 measures the airway pressure P.sub.aw and the probe 6 also measures the pressure P.sub.es in the esophagus.

(28) For example, the following lung-mechanical model equation is used in that case:
P.sub.es(t)=E.sub.cw*Vol(t)P.sub.mus(t)+P0+2.sub.n(5)
or also
P.sub.es(t)=E.sub.cw*Vol(t)P.sub.mus(t)+2.sub.n(6)
after the model equation (5) was differentiated according to time.

(29) The lung-mechanical factor E.sub.cw describes the elasticity on the basis of the chest wall (chest wall) of the patient P. The preferably normally distributed summand 2.sub.n describes, in turn, the process noise.

(30) In another embodiment, a model equation with additional summands, for example, the following lung-mechanical model equation, is preset and used instead of the model equations (1) or (4):
P.sub.aw(t)=R*Vol(t)+E*Vol(t)+I*Vol(t)+Q*(t)|Vol(t)|*Vol(t)+S*Vol(t)+P.sub.mus(t)+P0+3.sub.n.(7)

(31) Here, Q describes the resistance to the air flow, which the turbulent flow generates in a flexible tube from the ventilator 1 to the patient P and/or in the trachea of the patient P, S is the change in the compliance of the lungs and/or of the thorax as a function of the volume Vol of the lungs and I describes the resistance to the acceleration of the breathing air, this resistance I being negligibly low at a sufficiently low acceleration.

(32) The pneumatic parameter P.sub.mus to be determined is correlated, in addition, with the electrical respiratory signal Sig, which was generated from measured values of the sensor array with the measuring electrodes 2.1.1 through 2.2.2, or with a respiratory MMG signal. This correlation is described in the simplest case by the following model equation, which likewise belongs to the lung-mechanical model 20:
P.sub.mus(t)=k.sub.eff*Sig(t)+4.sub.n.(8)

(33) It is also possible that the respiratory signal Sig is generated by measured values from other sensors, for example, the probe 6 in the esophagus Sp or the gastric probe 14 in the stomach 14. Another model equation, which describes the relationship between P.sub.mus and Sig, is preferably used in this case.

(34) The factor k.sub.eff describes the neuromuscular efficiency of the respiratory muscles of the patient P, i.e., how well the respiratory muscles convert the electrical signals generated in the body of the patient P into pneumatic breathing activity. This factor k.sub.eff is an additional model parameter and is, as a rule, likewise variable over time. The preferably normally distributed summand 4.sub.n describes in turn the process noise.

(35) A respiratory signal Sig is generated in the embodiment described so far. It is also possible that two different respiratory signals Sig.sub.1 and Sig.sub.2 are generated, preferably with the use of measured values of different sensors, these two signals Sig.sub.1 and Sig.sub.2 being both correlated with the spontaneous breathing of the patient P.
P.sub.mus(t)=k.sub.eff.1*Sig.sub.1(t)+k.sub.eff.2*Sig.sub.2(t)+5.sub.n(9)
with two model parameters in the form of two factors k.sub.eff.1 and k.sub.eff.2.

(36) At least one of the model equations (1) through (7) as well as the model equation (8) or (9) form together a lung-mechanical model 20 of the exemplary embodiment which is stored in the memory 9. Each model equation of the lung-mechanical model 20 describes at least one relationship between the pneumatic parameter P.sub.mus to be determined and a plurality of measurable signals and is true for a respective scanning time t each. Because of the measurement noise (measured values are subject to errors) and of the process noise (the model is true only approximately), a plurality of model equations are used, even if this leads to a redundancy.

(37) In addition to the lung-mechanical model 20, a model 22 for the time curve of the pneumatic parameter P.sub.mus is used over a plurality of scanning times. Here, N is a preset number, which is preferably greater than or equal to 2.

(38) A preferred gradient model 22 for P.sub.mus comprises a state space model (state space model, process function) of the N.sup.th degree, i.e., a model of the form of
x(t.sub.k)=A*x(t.sub.k-1)+B*u(t.sub.k)+P(t.sub.k).(10)
Here, t.sub.k designates a vector with N scanning times, x(t.sub.k) designates a state vector with N components per occurring variable for the N different scanning times of t.sub.k, u(t.sub.k) designates an external input signal, e.g., a measured heart rate or a concentration of breathing gases, for example O.sub.2 and/or CO.sub.2, or of an anesthetic gas, and .sub.P(t.sub.k) designates the process noise, each doing so at the current scanning time t.sub.k. The value P.sub.mus(t.sub.k) to be determined for the pneumatic parameter P.sub.mus is a function of the state vector x(t.sub.k), wherein this function is also called observation model (observation function) and comprises the following observation equation:
P.sub.mus(t.sub.k)=g[x(t.sub.k)].(11)

(39) One embodiment of this gradient model 22 comprises an autoregressive model of the N.sup.th order, i.e.,
P.sub.mus(t.sub.k)=a.sub.1*P.sub.mus(t.sub.k-1)+a.sub.2*P.sub.mus(t.sub.k-2)+ . . . +a.sub.N*P.sub.mus(t.sub.k-N)+.sub.P(t.sub.k).(12)
Here, N is the number preset for the gradient model 22. The factors a.sub.1, a.sub.2, . . . , a.sub.N for the N previous values of P.sub.mus are preferably preset, for example, on the basis of a determination in a training phase prior to the use, but they may also be unknown and yet to be determined. The individual scanning times are designated by t.sub.k, t.sub.k-1, t.sub.k-N. The summand .sub.P(t.sub.k) describes the process noise and is preferably likewise treated as a normally distributed random variable with the expected value zero. The normally distributed random variables 1.sub.n, 2.sub.n, 3.sub.n, 4.sub.n and .sub.P may have different variances.

(40) The state vector has the form x(t.sub.k)=[P.sub.mus(t.sub.k), P.sub.mus(t.sub.k-1), . . . , P.sub.mus(t.sub.k-N)] in this embodiment. The function (11) (observation equation) has the form P.sub.mus(t.sub.k)=x.sub.1(t.sub.k).

(41) For example, it is assumed in the gradient model 22 that the pneumatic parameter P.sub.mus changes so slowly that the second derivation is negligibly small after time. The following gradient model with N=2 results from this:
P.sub.mus(t.sub.k)=P.sub.mus(t.sub.k-1)+[P.sub.mus(t.sub.k-1)P.sub.mus(t.sub.k-2)]+.sub.P=2*P.sub.mus(t.sub.k-1)P.sub.mus(t.sub.k-2)+.sub.P.(13)

(42) It is also possible to use an autoregressive gradient model with, e.g., N=2 and to estimate the two factors a1 and a2 in advance by means of a random sample and to set them thereby in advance. N may, of course, also be greater than or equal to 3.

(43) It is also possible to estimate the N parameters a.sub.1, a.sub.2, . . . , a.sub.N of the autoregressive model (12) at the operating time. One possibility of estimating these parameters is the following: The state vector x(t.sub.k) is expanded by the N unknown parameters. By calculating the expanded state vector, estimated values are determined for the N unknown parameters. A time curve model, for example, a random walk, is preferably preset for the unknown parameters. Even if the gradient model 22 with the state vector x(t.sub.k) was linear, the gradient model expanded by the N parameters is not, as a rule, linear. A nonlinear Kalman filter is used for the estimation.

(44) Breathing is a periodic process. The pneumatic parameter P.sub.mus being sought also changes therefore approximately periodically with a frequency that is, as a rule, variable over time. An oscillator model is used therefore as the gradient model 22 in one embodiment. The oscillator model may be, for example, a model of a damped harmonic oscillator, which [model] is discretized in time. In one embodiment, the phase and the amplitude of the spontaneous breathing, more precisely, the respective values at N respective scanning times of a breath, are at least two state variables of the state space model, which are correlated with the pneumatic parameter P.sub.mus being sought. Additional state variables of the state vector are, for example, estimated values of the pneumatic parameter P.sub.mus for previous scanning times.

(45) In a different embodiment, the course of the pneumatic parameter P.sub.mus within a breath is described as a weighted sum of a plurality of Gaussian bell curves (augmented Gaussian kernel). Each Gaussian bell curve is described by its respective height, its respective width and its respective position in time within the breath. These three variables, which are variable over time, form three state variables of the state space model, which are correlated with the pneumatic parameter P.sub.mus being sought. These are 3*M state variables in the case of M Gaussian bell curves. The weighting factors are either preset or are model parameters, whose values are estimated.

(46) It is also possible to preset and use a Gaussian process prior as the gradient model 22. Smooth, periodic or quasi-periodic signals can be modeled by means of a Gaussian process prior. The use of a Gaussian process prior makes it possible to introduce different hypotheses on P.sub.mus into the gradient model 22. A model for a Gaussian process is capable of depicting a plurality of different signal curves and can be readily adapted to different patients with different breathing efforts. Such a Gaussian process prior is described in Hartikainen, J. and Srkk, S.: Kalman filtering and smoothing solutions to temporal Gaussian process regression models, in: Proceedings of IEEE International Workshop on Machine Learning for Signal Processing (MLSP), 2010, which is incorporated herein by reference. It is possible to transform a Gaussian process prior into the form of a state space model and then to insert it directly into the gradient model 22. Since the Gaussian process prior is integrated into the state space model, it is not necessary, unlike in classical Gaussian processes, to invert large matrices.

(47) In one embodiment, the gradient model 22 comprises an input signal u(t.sub.k). Thanks to this input signal u(t.sub.k), the influence of external variables on the course of the pneumatic parameter P, for example, the influence of the airway pressure, of the lung volume, of the concentration of breathing gas or of the current heart rate of the patient, can be taken into account. The interaction between the patient P and a ventilator 1 can be readily taken into account in the gradient model 22 in this manner. This leads to more accurate predictions.

(48) The signal processing unit 5 carries out an initialization phase and a subsequent use phase. The signal processing unit 5 preferably applies the lung-mechanical model 20 in both phases, whereas it uses the gradient model 22 in the use phase only.

(49) The initialization phase comprises at least N scanning times, N being precisely the number preset for the gradient model 22. The signal processing unit 5 determines a respective value P.sub.mus,est(t.sub.i) for the pneumatic parameter P.sub.mus for these at least N scanning times t.sub.1, t.sub.2, . . . , t.sub.N during the initialization phase. The signal processing unit 5 preferably uses for this purpose the lung-mechanical model 20 as well as values of generated signals for the N scanning times t.sub.1, t.sub.2, . . . , t.sub.N. In another embodiment, an occlusion, during which the mechanical ventilation of the patient P is stopped for a short time period and the pneumatic parameter P.sub.mus can be measured directly, is carried out during the initialization phase. It is also possible that N values are available for the pneumatic parameter P.sub.mus from earlier determinations and are used or are preset in another manner.

(50) During the use phase, the signal processing unit 5 uses the gradient model 22 as well as the N previous values P.sub.mus,est(t.sub.i-N), . . . , P.sub.mus,est(t.sub.i-1), and, in addition, the lung-mechanical model 20 and signal values Vol(t.sub.i), Vol(t.sub.i), Sig(t.sub.i) in order to calculate an estimated value P.sub.mus,est(t) for the current scanning time t.sub.i.

(51) In one embodiment, the pneumatic parameter P.sub.mus is determined by means of a Kalman filter. In one embodiment of the exemplary embodiment, the Kalman filter describes a system, here the system with ventilator 1 and the respiratory system of the patient P, by means of a state space equation and of an observation equation. The state space equation results from the gradient model 22, and the observation equation [results] from the lung-mechanical model 20. In agreement with the usual notation, the abbreviating notation (k) is used instead of (t.sub.k).

(52) The gradient model 22 and the lung-mechanical model 20 are used simultaneously according to the present invention in order to determine the value of the pneumatic parameter P.sub.mus for a scanning time. In one embodiment, the gradient model 22 is implemented by means of a Kalman filter, the Kalman filter comprising for the gradient model 22 a state space model in the form of a state space equation
x(t.sub.k)=f[x(t.sub.k-1)]u(t.sub.k)+.sub.P(t.sub.k)(14)
and an observation equation
P.sub.mus(t.sub.k)=g[x(t.sub.k)].(15)

(53) The summand u(t.sub.k) takes into account the influence of measurable external factors at the N scanning times t.sub.k on the system described and hence on the pneumatic parameter P.sub.mus.

(54) The lung-mechanical model 20 is introduced via an additional observation equation.

(55) In one embodiment, the additional observation equation is obtained from the lung-mechanical model equations (1) and (8), where equation (8) is rearranged into:
Sig(t)=1/k.sub.eff*P.sub.mus(t)+5.sub.n(16)
with another summand 5.sub.n for the process noise, which is preferably likewise a normally distributed random variable with the expected value zero.

(56) The state space equation will now be:

(57) $\begin{array}{cc}\underset{\_}{x}\left(k+1\right)=\underset{\_}{F}\left(k\right)*\underset{\_}{x}\left(k\right)+{}_P& \left(17\right)\end{array}$$\begin{array}{cc}\mathrm{with}& \end{array}$$\begin{array}{cc}\underset{\_}{x}\left(k\right)=\left(\begin{array}{c}{P}_{\mathrm{mus},\mathrm{est}}\left(k-1\right)\\ P_{\mathrm{mus},\mathrm{est}}\left(k-2\right)\\ .Math..\\ P_{\mathrm{mus},\mathrm{est}}\left(k-N\right)\end{array}\right)& \left(18\right)\end{array}$$\begin{array}{cc}\underset{\_}{F}\left(k\right)=\left(\begin{array}{cccc}{a}_1& a_2& .Math.& a_N\\ 1& 0& .Math.& 0\\ .Math.& & & \\ 0& 0& & 1\end{array}\right)& \left(19\right)\end{array}$$\begin{array}{cc}\mathrm{and}& \end{array}$$\begin{array}{cc}{\underset{\_}{}}_p=\left(\begin{array}{c}{}_p\\ 0\\ .Math.\\ 0\end{array}\right).& \left(20\right)\end{array}$
The observation equation is

(58) $\begin{array}{cc}\underset{\_}{c}\left(k\right)=\underset{\_}{B}\left(k\right)+\underset{\_}{H}\left(k\right)*\underset{\_}{x}\left(k\right)+{\underset{\_}{}}_n\mathrm{with}& \left(21\right)\end{array}$$\begin{array}{cc}\underset{\_}{c}\left(k\right)=\left(\begin{array}{c}{P}_{\mathrm{aw}}\left(k\right)\\ \mathrm{Sig}\left(k\right)\end{array}\right)& \left(22\right)\end{array}$$\begin{array}{cc}\underset{\_}{B}\left(k\right)=\left(\begin{array}{ccc}R*{\mathrm{Vol}}^{}\left(k\right)& +E*\mathrm{Vol}\left(k\right)& +P0\\ 0& 0& 0\end{array}\right)& \left(23\right)\end{array}$$\begin{array}{cc}\underset{\_}{H}\left(k\right)=\left(\begin{array}{cccc}1& 0& .Math.& 0\\ 1/k_{\mathrm{eff}}& 0& .Math.& 0\end{array}\right)& \left(24\right)\end{array}$$\begin{array}{cc}{\underset{\_}{}}_n=\left(\begin{array}{c}{1}_n\\ 3_n\end{array}\right).& \left(25\right)\end{array}$
The state space vector x(k) is specified, in turn, by the equation (18).

(59) This embodiment has especially the following advantages: It is not necessary to detect individual breaths of the patient P or individual phases phases, especially inhalation and exhalation, during the breaths of the patient P and to adapt model parameters or other parameters of a model being used to the duration or the intensity of individual breaths. The measured values are rather obtained and processed uniformly and at a preset scanning frequency, i.e., independently from the breathing and from the ventilation. The lung-mechanical model 20 used contains two model equations (1) and (8). The model equation (1) describes a relationship between the pneumatic parameter P.sub.mus and a plurality of measurable pneumatic signals. The model equation (8) describes a relationship between the same pneumatic parameter P.sub.mus and a measurable electrical respiratory signal Sig. The separate application of these two equations (1) and (8) leads, as a rule, to different estimations for the pneumatic parameter P.sub.mus, namely, to a pneumatically determined estimation P.sub.mus,pneu(t.sub.i) and to an electrically determined estimation parameter P.sub.mus,EMG(t.sub.i) and P.sub.mus,MMG(t.sub.i) for the pneumatic parameter P.sub.mus. Both estimations refer to the same scanning time Thanks to the Kalman filter, an averaging is performed via these two estimations P.sub.mus,pneu and P.sub.mus,EMG, without weightings having to be preset for this averaging. The optional weightings 21 for the measured signals can be introduced without having to change the gradient model 22, namely, into the observation matrix B(k) of the observation equation (21). Standard programs, with which a Kalman filter can be rapidly implemented, are available commercially.

(60) Both models 22 and 20 are used simultaneously according to the present invention, e.g., as described by the equations (17) through (20) as well as (21) through (25). It is possible that no values at all or no sufficiently reliably measured values, e.g., no values for Vol(t.sub.x), for Vol(t.sub.x), for P.sub.aw(t.sub.x), for P.sub.aw(t.sub.x) or for Sig(t.sub.x), are available for an individual scanning time t.sub.x for at least one signal of the lung-mechanical model 20. Only the gradient model 22 is used in one embodiment for such a scanning time t.sub.x.

(61) It is also possible to use a separate parameter estimator 10 in order to estimate the values of the model parameters in a lung-mechanical model 20, in order to estimate them continuously. Individual breaths or the inhalation and the exhalation need not be detected and be distinguished from one another in this variant, either.

(62) The values of the model parameters are estimated from sets of signal values, with a plurality of sets of signal values being preferably used for a plurality of consecutive scanning times in order to estimate a set of model parameter values.

(63) The parameter estimator 10 preferably uses a statistical method to estimate the set of model parameter values. In one embodiment, a recursive method is used to estimate the parameter values. An updated model parameter value is estimated here using the previous model parameter values, the current signal value set and optionally at least one previous signal value set. In one embodiment, the parameter estimator 10 uses at least one of the methods Gradient-based Maximum Likelihood Estimation, Expectation Maximization, Variational Inference, Markov-Chain-Monte-Carlo, Ordinary Least Squares, Recursive Least Squares.

(64) It is also possible to use a separate Kalman filter for the parameter estimator 10. This Kalman filter uses, for example, the state space equation
x(k)=I*x(k1)+(Random Walk),(26)
in which

(65) $\begin{array}{cc}\underset{\_}{x}\left(k\right)=\left(\begin{array}{c}R\left(k\right)\\ E\left(k\right)\\ P0\left(k\right)\\ k_{\mathrm{eff}}\left(k\right)\end{array}\right)& \left(27\right)\end{array}$$\begin{array}{cc}\underset{\_}{I}=\left(\begin{array}{cccc}1& 0& 0& 0\\ 0& 1& 0& 0\\ 0& 0& 1& 0\\ 0& 0& 0& 1\end{array}\right)& \left(28\right)\end{array}$$\begin{array}{cc}\mathrm{and}\mathrm{wherein}& \end{array}$$\begin{array}{cc}\underset{\_}{}=\left(\begin{array}{c}1\\ 2\\ 3\\ 4\end{array}\right)& \left(29\right)\end{array}$
describes the respective error during the estimation of the four model parameter values R, E, P0, k.sub.eff. The model parameters are consequently variable over time. The observation equation is, for example,
P.sub.aw(k)=s(k)*x(k),+v,(30)
wherein
s(k)=(Vol(k),Vol(k),Sig(k),1)(31)
and wherein v is, in turn, treated as a normally distributed random variable with the expected value 0.

(66) It is possible to use a plurality of model parameters R.sub.1, R.sub.2 for the resistance, for example, a model parameter R.sub.1 for a linear dependence of the resistive pressure drop on the volume flow and another model parameter R.sub.2 for a square or other dependence of the resistive parameter R.sub.2 on the volume flow Vol, for example, the model equation (2). All model parameters R.sub.1, R.sub.2 are variable over time. The model parameter k.sub.eff for the neuromuscular efficiency may contain a plurality of individual model parameters k.sub.eff,1, k.sub.eff,2, . . . as well, for example, for different sets of measuring electrodes, which are attached at different locations on the skin of the patient P and/or measure electrical signals in a different way.

(67) In one variant, the state space equation

(68) $\begin{array}{cc}\underset{\_}{x}\left(k\right)=\underset{\_}{I}*\underset{\_}{x}\left(k-1\right)+\underset{\_}{},& \left(32\right)\end{array}$$\mathrm{in}\mathrm{which}$$\begin{array}{cc}\underset{\_}{x}\left(k\right)=\left(\begin{array}{c}{P}_{\mathrm{mus},\mathrm{est}}\left(k\right)\\ R\left(k\right)\\ E\left(k\right)\\ \mathrm{Po}\left(k\right)\\ k_{\mathrm{eff}}\left(k\right)\end{array}\right)& \left(33\right)\end{array}$
as well as the observation equation
c1(k)=B(k)*x(k)+err(k),(34)
in which

(69) $\begin{array}{cc}\underset{\_}{c1}\left(k\right)=\left(\begin{array}{c}{P}_{\mathrm{aw}}\left(k\right)\\ 0\end{array}\right)& \left(35\right)\end{array}$$\begin{array}{cc}\underset{\_}{B1}\left(k\right)=\left(\begin{array}{ccccc}1& -{\mathrm{Vol}}^{}\left(k\right)& -\mathrm{Vol}\left(k\right)& 0& -1\\ 1& 0& 0& \mathrm{Sig}\left(k\right)& 0\end{array}\right)& \left(36\right)\end{array}$$\begin{array}{cc}\underset{\_}{\mathrm{err}}\left(k\right)=\left(\begin{array}{c}\left(k\right)\\ \left(k\right)\end{array}\right)& \left(37\right)\end{array}$
are used, wherein is the process noise and is the measurement noise.

(70) FIG. 3 shows another variant. The parameter estimator 10 yields estimated values for the model parameters R, E, P0 and k.sub.eff, of which there are four in this case. An additional P.sub.mus estimator 13 located upstream of the P.sub.mus estimator 11 proper calculates a pneumatically determined estimation P.sub.mus,pneu(k) on the basis of the model equation (1), and, on the other hand, an electrically determined estimation P.sub.mus,EMG(k) on the basis of the model equation (8). The P.sub.mus estimator 11 proper calculates an aggregation, especially an averaging, via these two estimations determined in different ways. In one embodiment, the P.sub.mus estimator 11 proper uses an additional Kalman filter, which uses, for example, the following state space equation:
x(k+1)=I*x(k)+(Random Walk) with(38)

(71) $\begin{array}{cc}\underset{\_}{x}\left(k\right)=\left(\begin{array}{c}{P}_{\mathrm{mus},\mathrm{pneu}}\left(k-1\right)\\ P_{\mathrm{mus},\mathrm{EMG}}\left(k-1\right)\end{array}\right)& \left(39\right)\end{array}$$\begin{array}{cc}\underset{\_}{I}=\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right)& \left(40\right)\end{array}$
and

(72) $\begin{array}{cc}\underset{\_}{}=\left(\begin{array}{c}1\\ 2\end{array}\right)& \left(41\right)\end{array}$
wherein is a parameter for the measurement error: What error is inherently present in the two estimations P.sub.mus,pneu(k) and P.sub.mus,EMG(k) for the pneumatic parameter P.sub.mus?

(73) In one form of implementation, this embodiment leads to the following minimization task: Preset is the model equation P.sub.mus, wherein c describes the process noise (deviation between model and reality) and describes the measurement noise (measurement error during the generation of the signal Sig). For a scanning time window, k.sub.eff is considered to be constant over time. A signal value tuple{Sig(t.sub.i), P.sub.mus,est(t.sub.i)} is generated for a plurality of scanning times t.sub.i. For example, P.sub.mus,est(t.sub.i) is estimated with the use of the model equation (1). A best fit line Ag is then drawn by calculation through the signal value tuple. The sum of the distances between the signal value tuples and the best fit line Ag is minimized, doing so in a direction at right angles to the best fit line Ag and not in a direction at right angles to the x axis. FIG. 4 shows this in a two-dimensional coordinate system with Sig on the x axis and with the estimated parameter P.sub.mus on the y axis.

(74) The weightings 21, which the parameter estimator 10 uses, and the mode of operation of the weighting estimator 12 will be described in more detail below. In one embodiment, the sensor array with the electrodes (EMG sensors) 2.1.1 through 2.2.2 yield the electrical respiratory signal Sig. These sensors are arranged relatively close to the source for the electrical signals in the body of the patient P. Therefore, no weighting factors are used in one embodiment for the values of the signal Sig.

(75) The sensors 7, 4 and 6 measure pneumatic variables, and the sensors 7 and 4 and 6 are located at a relatively great distance from the source for the breathing activity in the body of the patient P. The following description pertains to an embodiment in which the sensors 2.2.1, . . . , 2.2.2 or 6 or 14, 7 and 4 yield measured values at each scanning time t, and the signal processing unit 5 generates from these measured values a triple with three signal values

(76) {Sig(t), Vol(t), Vol(t)}.

(77) It is also possible to generate a quadruple

(78) {Sig.sub.1(t), Sig.sub.2(t), Vol(t), Vol(t)}.

(79) For each signal value triple/quadruple and for each scanning time the weighting estimator 12 calculates a weighting factor (t) each. The parameter estimator 10 receives weighted signal values, e.g., each receives a triple

(80) {(.sub.i)*Sig(t.sub.i), (t.sub.i)*Vol(t.sub.i), (t.sub.i)Vol(t.sub.i).

(81) It is also possible to use different weightings for the signal values of a triple, e.g., {.sub.1*(t)*Sig(t), .sub.2*(t)*Vol(t), .sub.3*(t)*Vol(t)}, wherein the weighting factors .sub.1, .sub.2, .sub.3 are preset in advance and are constant over time.

(82) It is also possible to scale the variance of the summand .sub.n for the measurement noise inversely with the weighting factor: .sub.n,w(t)=.sub.n/(t). The weighting factor (t) is considered here to be variable over time.

(83) The following basic principle is preferably valis: The higher the information content, the higher is the weighting (t) of such a signal value triple. In one embodiment, the information content of a triple increases with decreasing frequency of occurrence of that triple in a random sample, i.e., in a set of signal value triples. In another embodiment, such triples in which one of the three signal values has a markedly greater relative amplitude than the other two signal values are provided with a high weight. The relative amplitude is defined, for example, as the deviation of the signal value from the arithmetic mean or from the median of all measured values for this signal.

(84) In one embodiment, a set of signal value triples is generated for a sequence of scanning times. An empirical density estimation is carried out with a set of signal value triples as the random sample in a three-dimensional space, for example, in a Cartesian coordinate system with the three coordinate axes Sig, Vol and Vol or in a polar coordinate space. For example, the three-dimensional space is divided into areas, e.g., cuboids, and a respective frequency is calculated for each area of the three-dimensional space. Each signal value triple in this area receives a weighting, which is, e.g., equal to the reciprocal value of the estimated frequency of the area.

(85) Freak values and obviously incorrect measured values are preferably removed automatically, doing so before the values for the model parameters are determined and before the weighting factors just described are calculated. Triples with freak values or with incorrect measured values are prevented hereby from receiving a high weight. Threshold values (limits) or ranges for measured values with a physiological and/or anthropological significance are preferably preset in order to detect freak values and incorrect measured values. It is also possible to recognize freak values and incorrect measured values by means of statistical methods. Triples with freak values and/or with incorrect measured values are sorted out or they receive the weighting factor zero or a very low weighting factor.

(86) While specific embodiments of the invention have been shown and described in detail to illustrate the application of the principles of the invention, it will be understood that the invention may be embodied otherwise without departing from such principles.

LIST OF REFERENCE NUMBERS

(87) TABLE-US-00001 1 Ventilator; it ventilates the patient P mechanically; comprises the signal processing unit 5 and the memory 9 2.1.1, First set of measuring electrodes on the skin of the 2.1.2 patient P; it provides measured values for the electrical respiratory signal Sig, which is correlated with the pneumatic parameter P.sub.mus, they are EMG sensors 2.2.1, Second set of measuring electrodes on the skin of the 2.2.2 patient P, closer to the diaphragm Zw; providing additional measured values for the electrical respiratory signal Sig, which is correlated with the pneumatic parameter P.sub.mus, they are EMG sensors 3 Connection piece with an image recording device and with an image processing unit; it measures the geometry of the body of the patient P in the thoracic region, from which the current filling level of the lungs, Vol, is derived 4 Optical sensor with an image recording device and with an image processing unit; it measures the geometry of the body of the patient P in the thoracic region, from which the current filling level of the lungs, Vol, is derived 5 Signal processing unit; it carries out the steps of the process according to the present invention; it has read access to the memory 9 6 Probe in the esophagus Sp; it measures the pneumatic pressure P.sub.es in the esophagus Sp; it is in fluid connection with the measuring catheter 8 7 Pneumatic sensor in front of the mouth of the patient P; it measures the volume flow Vol and the airway pressure P.sub.aw 7.1 Transducer of sensor 2; it taps air from the fluid connection between the lungs Lu of the patient P and the ventilator 1 7.2 Pressure sensor proper of the sensor 7; it receives measured values from the transducer 7.1 8 Measuring catheter in the esophagus Sp 9 Memory, in which the lung-mechanical model 20 and the gradient model 22 are stored and to which the signal processing unit 5 has read access 10 Parameter estimator; it has read access to the lung-mechanical model 20; provides values for the model parameters of the lung-mechanical model 20; it optionally uses weightings 21 11 P.sub.mus estimator proper; it has read access to the gradient model 22; receives estimated values for the model parameters of the lung-mechanical model 20; provides the estimated pneumatic parameter P.sub.mus 12 Weighting estimator; it provides weightings 21 as a function of frequencies of signal values 13 Upstream additional P.sub.mus estimator; it provides two different estimations P.sub.mus, pneu(k) and P.sub.mus, EMG(k) 14 Gastric probe in the stomach Ma of the patient P; it measures the gastric pressure P.sub.ga 20 Lung-mechanical model; it indicates at least one relationship between the pneumatic parameter P.sub.mus and a plurality of measurable signals for a respective scanning time; it preferably comprises a plurality of model equations with model parameters; it is stored in the memory 9 21 Weightings for the signal value sets; calculated by the weighting estimator 22 Gradient model for the pneumatic parameter P.sub.mus over a plurality of scanning times; stored in the memory 9 (t.sub.i) Weighting factor for a signal value set at the scanning time t.sub.i Ag Best fit line in a two-dimensional coordinate system with Sig as the x axis and P.sub.mus as the y-axis E Model parameter in the form of a lung-mechanical factor; elasticity of the lungs of the patient P k.sub.eff Neuromuscular efficiency of the respiratory muscles of the patient P; it is a model parameter Ma Stomach of the patient P; it receives the gastric probe 14 P Patient with the esophagus Sp, with the stomach Ma and with the diaphragm Zw; ventilated mechanically by means of the ventilator 1 from time to time P0 Model parameter in the form of a lung-mechanical summand: Residual pressure after complete exhalation by the patient P P.sub.aw Airway pressure, generated by a superimposition of the spontaneous breathing of the patient P to the mechanical ventilation by the ventilator 1; it is measured by the sensor 3 P.sub.es Pressure in the esophagus Sp of the patient P; it is measured with a probe 6 in the esophagus Sp P.sub.mus Pneumatic parameter to be determined for the spontaneous breathing of the patient P P.sub.mus, EMG(k) Estimated value for the pneumatic parameter P.sub.mus; determined by analysis of signals from the measuring electrodes 2.1.1 through 2.2.2 (EMG signal) P.sub.mus, pneu(k) Estimated value for the pneumatic parameter P.sub.mus, determined by analysis of signals from the pneumatic sensors 3, 4 and 6 P.sub.mus, est(t.sub.i) Approximately determined value of the pneumatic parameter P.sub.mus for the scanning time t.sub.i R Model parameter in the form of a lung-mechanical factor: Breathing resistance, which the airways of the patient P offer to the volume flow Vol Sig Electrical respiratory signal, generated from measured values from the measuring electrodes 2.1.1 through 2.2.2 Sp Esophagus of the patient P; it receives the optional probe 6 t.sub.k Vector with N scanning times Zw Diaphragm of the patient P