Multimodal method for detecting a change in a patient's physiological state and device for monitoring a patient for implementing such a method

Description

Technical field of the invention

[0001] The invention relates to methods for detecting a change in a patient's physiological state relative to a reference physiological state. The invention further relates to a patient monitoring device for implementing such a method.

[0002] The development of the invention is part of an effort to improve the characterization of respiratory discomfort. The invention thus offers a solution for improving the characterization of changes in physiological state, particularly changes in the respiratory state of a patient in anesthesia and intensive care. Technical background

[0003] The assessment of patients' respiratory status is routinely performed as part of a general diagnostic approach. This assessment is referred to as "preoperative" when conducted before the procedure, "intraoperative" when conducted during the procedure, and "postoperative" when conducted after the procedure. It is critically important for detecting any dyspnea that may occur during these different clinical phases. Of course, there are many other applications that require monitoring a patient's respiratory status, such as detecting breathing disorders during sleep.

[0004] Human respiratory function is autonomously regulated in the brainstem. However, when a patient's normal ventilation rate is disrupted, often resulting in respiratory distress, certain cortical regions may be activated. These cortical activations, responsible for a change in the patient's respiratory state, aim to recruit auxiliary respiratory muscles to compensate for any inadequate ventilation rate. This is referred to as "muscle recruitment" because muscle action is triggered by nerve signals associated with cortical activations. A disruption in the patient's normal ventilation rate therefore leads to a series of changes aimed at regulating various physiological constants, known as homeostatic changes, which can be measured by biomedical signals.

[0005] Examples include the presence of a specific signature in the patient's electroencephalogram (EEG), an increase in the electromyographic (EMG) activity of the patient's respiratory muscles, and / or a change in the dynamics of the respiratory tracings. It can also involve a change in the patient's cardiac activity—measured by electrocardiogram (ECG)—since a change in the patient's respiratory status can also trigger a stress response. Similarly, a change may be observed in the patient's electrodermal response.

[0006] Prior art is limited, at will, to focusing on the analysis and processing of only one type of physiological data (monomodal) to characterize changes in the patient's physiological state, or, when several types of physiological data (multimodal) are analyzed to characterize changes in the patient's physiological state, to exploiting them separately.

[0007] We know from the state of the art of many methods of detecting a change(s) in the physiological state of the patient which are based only on the analysis and exploitation of a single type of physiological signal.

[0008] Such methods are described, for example, in documents US5820560 A, US2010252038 A1, and WO2013164462 A1. The described methods are based, depending on the case, solely on data related to the electromyographic activity of the patient's respiratory muscles or on the patient's electroencephalographic data. The single-modal methods (based on information from a single type of signal) described in these documents require a step to confirm that a change in the observed signal corresponds to a real change in the patient's physiological state, according to the measured signal. This requires the ability to first associate the timing of a physiological change in the measured signal(s) with the physiological event itself, given that the measured signal(s) in question may be very noisy.The aforementioned confirmation step may therefore require the use of several different sensors or other means of measurement for the analysis of the same type of homeostatic change in order to more precisely identify the time of the physiological event.

[0009] Furthermore, still with the aim of confirming a real change in the patient's physiological state, it may be necessary, using prior art methods, to apply complex treatments to the recorded signals.

[0010] This is the case, for example, in document US5820560 A, where a series of processing steps are required to extract relevant information from electromyograms. This document itself (Col. 5, I. 51 et seq.) highlights the difficulty in maintaining the noise level in the signals as low and constant as possible. Signal amplifiers and signal converters are therefore used to compensate for the effects of diaphragm movement. Numerous sensors are also required.

[0011] The solutions proposed in these single-modal methods do not sufficiently limit the occurrence of false alarms from the respiratory support device and / or are too complex to implement. Furthermore, another drawback of single-modal methods is that, inherently, they only allow observation of one physiological parameter of the patient, which does not allow for characterization of the patient's overall condition.

[0012] Multimodal methods combining several physiological signals for the detection of a change in the patient's physiological state have also been proposed.

[0013] An example of such a method is described in document WO 2013 / 140229 A1. This document discloses a method for regulating a mechanical respiratory support device. According to the approach used in this document, the patient's respiratory support is regulated based on the evolution of the Paw / Eadi ratio, where the parameter Paw is the value of the signal due to the exerted ventilatory pressure, while the parameter Eadi is the value of the electrical signal caused by the patient's diaphragmatic activity. The parameter Paw represents the patient's spontaneous respiratory activity and is thus associated with the muscle pressure Pmusc. The parameter Eadi represents the neuronal respiratory activity that causes diaphragmatic movements. The ventilatory support provided to the patient is regulated based on the Paw / Eadi ratio.This ratio allows us to deduce the proportion to which the patient contributes to respiration relative to the total respiration generated by the patient and the ventilatory support device. It therefore provides an estimate of the extent to which the patient intervenes in respiration to adjust the respiratory support. The extent to which the patient contributes to respiration does not characterize a change in their physiological state. Indeed, a number of respiratory movements by the patient may originate from the patient themselves without being related to respiratory distress. Furthermore, this multimodal method does not utilize the correlations between the two aforementioned physiological parameters but extracts a single piece of information from the measurements of the physiological signals associated with these two parameters.

[0014] US patent application 2004 / 0254493 describes a method for detecting changes in a patient's physiological state based on the analysis of multiple physiological signals. This document focuses more specifically on the detection of breathing disorders occurring during a patient's sleep. The detection method is based on analyzing changes in electroencephalographic signals during the individual's respiratory cycle. In addition to electroencephalogram analysis, this method also incorporates other signals characterizing the patient's physiological state, such as electrooculograms, electromyograms, and measurements of oral and nasal airflow. However, the method is limited to the simultaneous recording of the various signals without any processing to take advantage of the potential complementarities arising from the combined acquisition of these different signals.Indeed, only the data from electroencephalographic signals undergoes specific processing. The practitioner therefore receives raw data that they must interpret themselves. Consequently, the method is not suitable for real-time adjustments to the patient's respiratory support. This likely stems from the fact that monitoring the respiratory status of patients with sleep-related disorders generally aims to establish a diagnosis, with treatment occurring afterward. Summary of the invention

[0015] The invention overcomes the drawbacks of known prior art methods by exploiting and correlating data from various physiological signals characterizing the patient's condition in order to better detect and characterize changes in the patient's condition. The invention is defined by claims 1 to 12.

[0016] In this respect, the invention relates to a method for detecting a change in a patient's physiological state relative to a reference physiological state associated with a reference matrix X ref X signals ref,i (i = [1...N+1], N integer) and at a reference period W 0 , the process implementing in a loop the following steps: A) perform in M ​​time segments m of an observation window the measurements of the patient's electroencephalographic S1 signals according n roads and to p instants to generate M measure matrices X 1,m (m ε [1...M], M integer) comprising n*p samples and simultaneously, perform measurements of at least N other physiological signals S N+1 (N ≥ 1) of the patient, different from the electroencephalographic signals S 1, to generate N other measurement matrices X i =2 ... N + 1,m, C) for each time segment m of the reference period W 0 , determine distances d i (m) between each X signal i,m measured and the X component ref,i of the reference period, D) transform the distances d i (m) calculated in step C) via a log function d(m), E) choose the precision of the values ​​obtained at the end of step D by selecting a reference quantile u 0 on all values ​​of the reference period W 0 and associate with each distance d i (m) a scalar variable pi=1,..., N+1 (0 < p i=1,..., N+1 < 1) dependent on the reference quantile u F) merge the data obtained at the end of step E) by performing a weighted sum of the variables pi=1,..., N+1 to obtain resulting distances of merger (m) , G) determine a deviation e(m) from the reference physiological state as a function of the distances d merger (m).

[0017] The invention thus proposes a multimodal method for characterizing a patient's physiological state, in which data from different types of measured physiological signals are processed and fused to deduce a change in the patient's physiological state. The invention therefore goes beyond prior art methods and makes it possible to obtain simplified variables resulting from the combination of data acquired from the different signals. The information obtained is therefore more qualitative, more reliable, and more precise compared to what is possible with the prior art. The analysis of different physiological signals S1, ..., Sn+1 allows for a better characterization of the patient's state.The fusion of data from these different signals makes it possible to reduce, or even eliminate, non-qualitative data segments, minimize the impact of artifacts on the final data, and thus obtain higher-quality information and reduce false alarms in the medical assistance device. The invention therefore offers a real-time monitoring solution that automatically controls the respiratory assistance device and reduces unnecessary interventions by the practitioner on the patient.

[0018] Depending on various characteristics of the invention, which may be considered together or separately: In step B), the electroencephalographic signals S1 are centered and filtered into Q predetermined frequency bands to obtain MxQ matrices of filtered measurements X 1,m,q (q ε [1...Q]) and we determine MxQ normalized spatial covariance matrices by the formula: C m , q = X 1 , m , q X 1 , m , q T trace X 1 , m , q X 1 , m , q T ; In step C), Riemannian distances d are determined i,r (m) between each normalized spatial covariance matrix C m,q and the X components ref,i of the reference matrix associated with electroencephalographic measurements; during step C), statistical distances are determined i,s (m) between one (or more) signal(s) X i,m measured(s), other than electroencephalographic signals, and the X component(s) ref,i of the reference period associated with it(s); d merger (m) is calculated according to the following relationship: d fusion m = ∑ i λ i p i = 1 , … , N + 1 m , with p i = 1 , … , N + 1 m = 1 1 + e u 0 − log d i = 1 , … , N + 1 m where λ i is the weight assigned to each variable p i=1,...N+1 , the weights λ i satisfying ∑ i λ i = 1, u0 is the reference quantile, d 1,q (m) are the Riemannian distances and d i>1 (m) are the statistical distances; the weights λ i are selected according to the quality of the signal S i to which they are associated or according to the importance of the signal S i to which they are associated in determining a change in physiological state in the patient; a filtering step of the resulting distances d merger (m) is implemented prior to step F); the filtering consists of smoothing obtained by a moving average over the last L segments of the observation window (1 <L<M) appliquée aux distances résultantes d merger (m) , said distance d merger (m) is then calculated according to the following relationship: d fusion L m = 1 L ∑ i = 0 L − 1 d fusion m − i the last L segments are chosen so that the moving average is calculated taking into account the signals measured only with a delay of between 10 seconds and 1 minute, preferably between 30 seconds and 1 minute from the beginning of the observation window; said other physiological signals S i are cardiac activity and / or respiratory activity and / or muscular activity and / or a signal related to movement; the method further comprising a step in which each of the deviations e(m) is compared to a determined threshold Θ.

[0019] The invention further relates to a patient monitoring device for implementing a process as previously described, the device comprising: means for measuring electroencephalographic signals to measure the patient's electroencephalographic signals, means for measuring cardiac activity and / or respiratory activity and / or muscular activity and / or movement, means for real-time signal processing, said means including at least means for determining the X components i,ref of the reference matrix, means of calculating deviations e(m) from the reference situation and means of merging the data.

[0020] The invention finally relates to a medical assistance device comprising a monitoring device as previously described. Brief description of the figures

[0021] Other objects and features of the invention will become clearer in the following description, made with reference to the accompanying figures, in which: There figure 1is a schematic diagram illustrating certain steps of the detection process according to the invention, from the extraction of data from various physiological signals to the fusion of said data; The figure 2a illustrates the performance of three detectors for a SU1 subject: the first figure from the left illustrates the performance (AUC) of an independent detector using only S1 electroencephalographic signals, the second figure from the left illustrates the performance (AUC) of an independent detector using only S2 respiratory signals, and the third figure from the left illustrates the performance (AUC) of a detector using fused fusion distances; the fourth figure from the left (and therefore the rightmost figure) illustrates the ROC curves of these three detectors; The figure 2b differs from figure 2a in that the performances are illustrated for a SU2 subject; The figure 3illustrates the effects of a filtering (smoothing) step on the performance of different detectors, including an independent detector using respiratory S2 signals alone, an independent detector using electroencephalographic S1 signals alone, and a detector using fused fusion distances (in that order when reading from left to right). Detailed description of the invention Description of the general mode of the detection method according to the invention

[0022] The invention relates to a method for detecting a change in a patient's physiological state. The patient's physiological state is characterized, in particular, by a set of functions and reactions, also called physiological parameters. Hereafter, a distinction will be made between a reference physiological state and a modified physiological state, the latter characterizing a change in the patient's physiological state relative to their reference physiological state. A change in the patient's physiological state is considered a change in their physiological state relative to a reference physiological state when this change is due to at least one homeostatic modification, as defined in the following paragraphs.

[0023] The reference physiological state is not necessarily the patient's physiological state when healthy, that is, the physiological state in which the patient finds themselves when their functions and reactions are normal and / or when the patient has no pathology. The reference physiological state is the patient's observed physiological state at a given moment, whether the patient is in good or poor health. This observation is associated with a number of measurements, which are themselves associated with signals that will be described in more detail later.

[0024] The patient's altered physiological state is not necessarily the patient's physiological state when ill, that is, the physiological state in which the patient finds themselves when their functions and reactions are abnormal and / or when the patient presents with one or more pathologies. An altered physiological state of the patient corresponds to a physiological state that diverges from the reference physiological state. This occurs when at least one, or in practice several, of the patient's functions or reactions deviate from the functions and / or reactions the patient exhibits in the reference physiological state. In this case, it is referred to as a homeostatic alteration.

[0025] A change in physiological state therefore does not simply consist of a healthy patient becoming ill or vice versa, but rather in any change that causes them to move from their reference physiological state to another state following a homeostatic modification. In other words, within the framework of the invention, the patient's physiological state does not imply whether or not they are healthy.

[0026] In the context of the invention, the reference physiological state is associated with a matrix X ref, called the reference matrix, constructed from a plurality of components X ref,i(i = [1...N+1], N integer). Each component Xref,i of the reference matrix Xref is associated with a measurement of a given dimension, whose elements are the different physiological signals of the patient at the reference physiological state. This reference physiological state is associated with a reference period W0. The dimensions of the reference matrix therefore depend on: 1) the number of physiological signals measured, regardless of the type of physiological signal to which the measured signal is actually assigned, and 2) the number of times at which the measurements of the physiological signals are subsequently performed, that is, the measurements performed outside the reference physiological state, which itself is associated only with the reference period W0.

[0027] Regarding point 1) above, an Xref,i component of the reference matrix can be associated with spatiotemporal measurements, particularly when assessing the physiological parameter requires measurements to be taken in different regions of the body or a specific area of ​​the patient's body over time (e.g., different electroencephalographic signals or different muscle signals recorded simultaneously). Such measurements are frequently performed when assessing the patient's cortical or muscular activity. They involve a series of electrodes, and sometimes needles, placed in appropriate areas of the patient's body to map, and thus spatially represent, the evolution of the physiological parameter in question.For example, electroencephalographic measurements using 10 electrodes are associated with 10 different physiological signals, with each electrode assigned a physiological signal. Furthermore, if these measurements are performed at 8 different times, regardless of any segmentation, the Xref,i component of the reference matrix associated with the electroencephalographic measurements is a 10 x 8 matrix. However, this does not presuppose the dimensions for the other components of the reference matrix Xref,i, which also includes other Xref,i signals.

[0028] With regard to point 2) above, it is important to recall that while the invention seeks to establish correlations between different types of homeostatic changes, and consequently it is essential, as will be seen below, that measurements of the different types of physiological signals be carried out in parallel, i.e., simultaneously, or at least over a common time window, this does not mean that the time window over which the measurements are taken must be of the same length (number of samples or segments) for all the physiological parameters measured. The practitioner may indeed wish to observe the evolution of one or more physiological parameters for a longer period than others. The invention remains within its scope as soon as the method is implemented with at least two types of physiological signals.

[0029] The reference matrix X ref comprises, in any event, as many components X ref,i as there are types of physiological signals observed. In the invention, there are ai = [1...N+1] types of physiological signals measured.

[0030] In a first step A) of the detection method according to the invention, electroencephalographic measurements of the patient are performed in M ​​time segments of an observation window. Electroencephalographic (EEG) measurements allow for the measurement of the electrical activity of the patient's brain and are presented in the form of recordings.

[0031] The observation window is a temporal observation window. It defines a time interval during which measurements are taken at times whose frequency can be predetermined. The M time segments define sub-intervals of the observation window during which measurements of the electroencephalographic S1 signals are taken continuously or discretely. Each of the M time segments is assigned a duration denoted W. If the observation window is the reference period defined previously, each segment of the reference period is similarly assigned a duration denoted W0,m.

[0032] We define p instants at which electroencephalographic measurements are taken within a time segment m of the observation window. We further define n channels through which the patient's electroencephalographic measurements are performed. In practice, the patient wears a headset with n electrodes positioned appropriately on the patient's head to measure the electroencephalographic signals S1, each electrode being associated with one channel. For each time segment M, we thus generate a matrix of dimension nxp, the elements of the matrix corresponding to spatiotemporal measurements. We will denote by X1,m, m ∈ [1...M], and M an integer, the M measurement matrices thus generated.

[0033] Simultaneously, still during the first step A) of the method according to the invention, measurements are taken of at least one other type of physiological signal from the patient. Indeed, as previously stated, the method according to the invention is a multimodal method for detecting a change in physiological state. As such, the method according to the invention is based on the simultaneous analysis of several physiological parameters of the patient. It is sufficient that at least one other type of physiological parameter, other than cortical activity (EEG), is observed for the method to be considered a multimodal method. The other physiological signal(s) measured are electromyograms, respiratory flow rate, chest distensions or respiratory sounds, electrocardiograms (ECGs), electrodermal measurements, or motion measurements.This is not in any way limiting and other physiological parameters useful to the person skilled in the art to characterize the physiological state of the patient can be measured.

[0034] Within the scope of the invention, the other types of physiological signals, N+1, N ≧ 1, measured other than the electroencephalographic S1 signals, are defined as N+1. Therefore, there are N+1 types of physiological signals from the patient that are measured simultaneously over at least one common observation window. Similar to the electroencephalographic S1 signals, measurements of the other physiological signal(s) of the patient are taken in the M time segments of the observation window. This generates, for each time segment m, N other measurement matrices Xi = 2...N+1,m. The dimensions of each matrix XN,m depend on the type of physiological parameter considered.

[0035] Substeps B1), B2), and C1) described below relate to processing applied to electroencephalographic S1 signals according to a particular, i.e., non-limiting, implementation of the present invention. This implementation concerns the case where the electroencephalographic S1 signals are processed in Riemannian geometry. Reference may be made, in particular, to document WO 2013 / 164462 A1, which describes such steps. It should be noted that any other measured signal, i.e., other than electroencephalographic S1 signals, for which this processing proves more appropriate, may also be subjected to the processing described in these steps.

[0036] According to a particular implementation, in an optional second step B), and more precisely a first substep B1), each signal of the matrix X1,m is centered and filtered in Q predetermined frequency bands to obtain MxQ filtered measurement matrices X1,m,q, q ∈ [1...Q]. It is recalled that the measurement matrices X1,m were obtained by electroencephalographic measurements. The sequential filtering and centering operations of the measurement matrices have the usual mathematical meaning given to this type of operation. Advantageously, it should be noted that the measurement matrices X1,m are centered and filtered in five frequency bands corresponding to the frequencies conventionally used for electroencephalography, namely 1-4 Hz, 4-8 Hz, 8-12 Hz, 12-24 Hz, and 24-48 Hz.This allows us to retain only the information from the relevant brain regions of the patient, with each frequency range associated with one or more types of patient movement. For example, in the case of multiple electromyograms measured at different locations on the diaphragm, it would have been appropriate to select frequency bands specific to the diaphragm's muscular dynamics.

[0037] Also during the second optional step B), a second sub-step B2) consists of determining normalized spatial covariance matrices from the following formula: C m , q = X 1 , m , q X 1 , m , q T trace X 1 , m , q X 1 , m , q T where Cm,q are the normalized spatial covariance matrices, X1,m,q are the filtered measure matrices, and X1,m,qT are the corresponding transposed matrices. The spatial covariance matrices Cm,q characterize the synchronization of cortical activities over time. They include, within the diagonal elements, local synchronizations, and, outside the diagonals, long-range synchronizations, which are then characteristic of the dynamics of the neural network.

[0038] According to a particular implementation, during a third step C), more precisely a first substep C1) of the third step, Riemannian distances d are determined for each time segment m of the reference period W0. 1,r (m) between each normalized spatial covariance matrix C m,qand the Xref,1 component of the reference matrix associated with the electroencephalographic measurements. More precisely, for each time segment m of the reference period W0, we calculate the Riemannian distance d 1,r (m) between each normalized spatial covariance matrix C m,q calculated in the time segment m, of duration W, at time t: t + W and the component Xref,1 of the reference matrix calculated in the reference period W0. In this respect, in each frequency band, prototype matrices PRq,r, where r ∈ [1...R], are determined from the distribution of the spatial covariance matrices. C m,q . Each prototype represents a subclass of the synchronization, and is estimated here by a Karcher mean of the reference spatial covariance matrices C m,qof neighborhood. It should be noted that in the case of a single prototype, this corresponds to the average covariance matrix for the entire reference period.

[0039] A procedure for calculating PR prototypes q,r according to a specific, and therefore non-limiting, implementation is described below and is applied for each frequency band. According to this particular implementation, the calculation relies on the dynamic cloud algorithm (E. Diday, "A New Method in Automatic Classification and Pattern Recognition: The Dynamic Cloud Method," Revue de Statistique Appliquée, Vol. 19-(1971) no. 2, pp. 19-33) adapted to the Riemannian metric. It should be noted that in signal processing, the classical Frobenius norm is usually used to define distances between covariance matrices (which are, by definition, positive-definite Hermitian matrices). This approach assumes a normed vector space of zero curvature. However, the space of positive-definite Hermitian matrices is more accurately described as a metric space with negative curvature.This particular implementation preferably uses the tools of Riemannian geometry to manipulate covariance matrices. In this framework, the distance between two matrices corresponds to the geodesic in the space generated by their Hermitian property, and the average of the covariance matrices no longer corresponds to an arithmetic mean as classically, but to a geometric mean.

[0040] According to this particular implementation, the distance dist, i.e. called d i,r The distance between the covariance matrices is the following Riemannian distance: if P1 and P2 are two matrices, then: dist P 1 P 2 = ∑ k = 1 K ln 2 λ k 1 / 2 Or dist (P1, P2) is the Riemannian distance between the prototype matrices P1 and P2, and the λk are the K eigenvalues ​​of the joint matrix P 1 − 1 P 2 .

[0041] The Riemannian distance dist checks the three properties of a distance, namely symmetry, separation, and triangle inequality.

[0042] According to a particular implementation, the mean can be calculated using a gradient descent procedure that converges rapidly (Pennec, Statistical computing on manifolds for computational anatomy. 2006. Doctoral thesis. University of Nice Sophia Antipolis 2006): where the operators check: exp CM W = CM 1 / 2 exp CM − 1 2 W CM − 1 2 CM 1 2 log CM W = CM 1 / 2 log CM − 1 2 W CM − 1 2 CM 1 2

[0043] Subsequently or simultaneously, during a second substep C2) of the third step C, statistical distances are calculated. i,s (M) between one (or more) signal(s) X i,m measured(s) and the component(s) X ref,i of the reference period W0. More precisely, for each time segment m, we calculate the statistical distances d i,s (m) between one (or more) signal(s) X i,mmeasured in the time segment m, of duration W, at time t: t + W and the component X ref,i of the reference period W0. The signal(s) in question is / are those that have not been processed in Riemannian geometry for whatever reason, for example because Riemannian geometry is not suitable or proves unsuitable. In the case of this particular implementation, a distance di therefore corresponds either to a Riemannian distance di,r (m) or to a statistical distance d i,s (m) taking into account the S1 electroencephalographic signals and other signals. The calculation of statistical distances is carried out according to the classical mathematical treatment.

[0044] It should be noted that steps B) and C) can be implemented without using measurements of the S1 electroencephalographic signals in Riemannian geometry. Indeed, using S1 electroencephalographic signals in Riemannian geometry is easier because it facilitates measurements taken at different points on the patient's head. However, this is not mandatory, even though this method yields better results. An alternative to the Riemannian distance between covariance matrices could be the Euclidean distance between matrices (the Frobenius norm of the algebraic difference of two matrices). Another alternative would be to directly measure distances between EEG signals, without necessarily using covariance matrices. For example, one could use Hellinger statistical distances or the Bhattacharyya distance (Basseville, M. (1989)).Distance measures for signal processing and pattern recognition. Signal processing, 18(4), 349-369.).

[0045] Other methods described in documents FR 2 903 314 A and US 2004 / 0254493 A1, already cited in the preamble to this description, also allow for the exploitation of such signals. Document FR 2 903 314 A describes another method for exploiting electroencephalographic signals in which distances correspond to an electroencephalographic potential. In document US 2004 / 0254493 A1, distances correspond to the differences between a maximum EEG segment power and a minimum EEG segment power throughout the respiratory cycle. This is important because the acquisition of S1 electroencephalographic signals may not be available or may contain numerous artifacts.

[0046] More generally, distances can be calculated by any other method known in the prior art. In any case, a "distance" quantifies the statistical difference between two values ​​of a physiological parameter. In other words, the distance is the mathematical expression of the change in a homeostatic constant at a given instant and its value over the reference period, and can be associated with the occurrence of a physiological change in the patient. To determine the merged distances of merger (m) , As described later in this description, it suffices to determine the distances between each matrix of measurements taken in the time segments m and the component X ref,i (i = [1 ... N+1]) of the associated reference matrix. Let us consider only the signal S2 of the figure 1of this application. A classifier under consideration calculates the distance of a variable / value measured at a given time and the distribution of the same variable / value calculated over the reference period. More specifically, it could be a Support Vector Machine (SVM) classifier to a class or a classifier based on the Mahalanobis distance. Therefore, the classifiers used to calculate distances are not limited.

[0047] The invention therefore does not lie in the methods used to calculate the distances, nor even in the use of distances to merge the data from the different measurements, but in the act of merging the data itself, as will be seen in more detail later. Thus, step C) consists of determining distances d i (m) between each X signal i,m measured and component X ref,iof the reference period, for each time segment m of the reference period W 0 , the distances d i (m) which can be Riemannian distances di,r (m) and / or statistical distances di,s (m) and / or any other type of distance, for example of the type seen previously.

[0048] We now describe, with reference to the figure 1 This is a specific implementation that fuses data obtained from Riemannian and statistical distances. The resulting data fusion makes the detection of physiological changes more qualitative, reliable, and precise. Furthermore, this fusion allows for a better characterization of the patient's physiological state.

[0049] According to a particular implementation, during a fourth step D) of the process, the distances d are transformed i (m) calculated in step C) via a log function d(m). The goal of this step is to transform the Riemannian and statistical distances to stabilize their variance and ensure their values ​​are between 0 and 1. Indeed, significant differences can be observed between the distances calculated within the different time segments m, regardless of the type of distance considered, i.e., statistical or Riemannian distance. It is therefore necessary to better mathematically represent the relative variations between the different distances calculated in the M time segments m, for each of the Riemannian and statistical distances. In a preferred embodiment, we aim to make the distributions closer to the Gaussian distribution.However, the use of a Gaussian distribution is not mandatory, and other types of distributions within the grasp of a person skilled in the art can be considered insofar as they allow for the stabilization of the variances of the measured data. At the end of step D), we therefore obtain the logarithms log(di(m)) of the aforementioned distances.

[0050] According to this particular implementation, during a fifth step E), specifically a first substep E1) of the fifth step E), the precision of the values ​​obtained at the end of step D) is chosen by selecting a reference quantile u 0 on the set of values ​​of the reference period W 0. In other words, during this first sub-step E1), we choose the reference quantile u 0 which will be applied to the set of values ​​measured during step D) by selecting it on the set of values ​​of the reference period W 0.

[0051] In practice, the appropriate quantile is chosen based on the number of points in the distribution and the desired precision for describing the distribution, that is, for describing all the values ​​measured in the reference period W0. Other criteria, such as the number of outliers or the number of available data points, can also be taken into account to determine the reference quantile u0. On an empirical basis, the present inventors have demonstrated the relevance of the ninth decile, namely the 90% quantile, as the reference quantile u0 for obtaining the desired precision while taking into account the parameters of the distribution, in this case a Gaussian distribution (size, outliers, etc.).Therefore, selecting the reference quantile u0 does not necessarily require additional analysis or calculation steps, since it is perfectly possible to implement the first substep (E1) with the aforementioned empirical value. In other words, the reference quantile u0 can be predetermined / preselected. For the sake of completeness, the first substep (E1) could be implemented before step E), provided the reference period data (W0) are known.

[0052] According to this particular implementation, in a second sub-step E2), each distance is associated with i (m) a scalar variable pi=1,..., N+1 (0 <p i=1 ..., N+1 <1) depending on the reference quantile uo .In other words, during this second substep (E2), the quantile u0, pre-selected or selected during the first substep (E1), is used to perform scalar assignments. Indeed, each of the log(di(m)) calculated at the end of step D) is assigned a scalar variable. pf=1 , ...,N + 1 according to the (pre)selected reference quantile u0. This step ensures that all data associated with the observed physiological parameters can be classified relative to one another, regardless of the type of data to which they were originally linked, grouped by type, and merged (particularly during step F). At the end of step E, we therefore have a classifier in which the data obtained from the different measured physiological signals can be compared, classified, and merged.

[0053] According to a preferred embodiment, the scalar variables pi=1,..., N+1 (0 <pi =1,...,N+1 < 1) are calculated from the following transformation: p i = 1 , … , N + 1 m = 1 1 + e u 0 − log d i = 1 , … , N + 1 m Or pi=1,...,N+1 are the scalar variables, u0 is the reference quantile, d i=1 ... N+1 (m) are Riemannian distances and statistical distances.

[0054] By being calculated in this way, the scalar variables pf=1 , ...,N + 1 therefore take the form of probabilities. The use of scalar variables pi=1,..., N+1 in such a form makes it easier to process mathematically and therefore simplifies the implementation of the detection process according to this particular implementation.

[0055] In a sixth step F), the data obtained at the end of step E) are merged by performing a weighted sum of the variables p i=1,...,N+1 to obtain resulting distances of merger (m). In other words, for each of the M time segments m, we obtain a distance d merger(m) resulting from the merging of the data obtained at the end of step E) affected by a weight, i.e. a weighting coefficient.

[0056] It should be noted that one advantage of segmenting the observation window into M time segments m is to preserve a good correlation of data from different types of physiological signals.

[0057] Furthermore, while data fusion itself reduces data and thus provides greater algorithmic efficiency, it is also highly effective in that it allows for the very precise deduction of changes in the patient's physiological state. Indeed, each fused distance of merger(m) takes into account the different types of physiological signals measured and allows the integration of the information contained in these different signals. Instead of superimposing measurements, possibly from different sensors, of the same type of physiological signal as was done in the prior art, the aim is to superimpose and couple measurements of different types of physiological signals. In the context of the invention, "fusing" the measurements does not refer to simply pairing the measurements, that is, considering them in pairs, trios, etc. "Fusing or combining" should be understood as generating a result, here of merger (m) , by the mathematical transformation of the data from the measurements, the data being the distances of i (m) previously calculated. Any detection of a change in the patient's physiological state is therefore the result of a combination / fusion of data from the various physiological parameters. A change in the patient's physiological state established on the basis of the method according to the invention means that several homeostatic modifications, at least two, have therefore taken place. It is thus more representative of the patient's overall condition.

[0058] Performing weighted sums of scalar variables p i=1,...,N+1 (0 < p i = 1,...,N+1 < 1) , that is, to sum the scalar variables p i = 1,...,N+1 , previously assigned a weighting coefficient, allows for a substantial improvement in the accuracy of the merged distances. merger(m) Indeed, it is possible to assign less weight to data from the most contaminated signals, for example by noise, artifacts, etc., and conversely to assign more weight to signals of better quality. In this case, the weighting coefficient is defined based on a post-hoc criterion.

[0059] It is also possible to define a weighting coefficient based on an a priori criterion. For example, it is possible to assign a weighting coefficient according to the importance of the signal Si to which it is associated in determining the change in the patient's physiological state; that is, according to the importance of the observed physiological parameter in the change in the patient's physiological state. An a priori criterion has the advantage of objectifying the calculation of d merger(m) in comparison with an a posteriori criterion. However, nothing prevents the use of a weighting coefficient resulting from an a priori criterion and an a posteriori criterion. The type of weighting coefficient selected is not limiting within the scope of the present invention. Moreover, it is possible to assign the same weighting coefficient to all distances of the scalar variables p i=1,...,N+1 calculated, which amounts to giving equal importance to all the measured signals.

[0060] All the advantages previously mentioned contribute to improving the detection of changes in the patient's physiological state. The detection method according to the invention is therefore more reliable and efficient compared to methods known in the prior art.

[0061] According to a preferred embodiment of the present invention, d merger (m) is calculated as follows: d fusion m = ∑ i λ i p i = 1 , … , N + 1 m where λ i is the weight assigned to each variable p i=1,...,N+1 . Regarding scalar variables p i=1,...,N+1 , They can, for example, be calculated using formula (3) described above. The aforementioned advantages are not reiterated. However, it is specified that it is preferable for the sum of the weighting coefficients λi to be such that λ1 + λ2 + ... + λN = 1. This aims to simplify the mathematical processing of the data in this particular implementation of the detection method according to the invention.

[0062] Once the distances of merger (m) calculated, we determine a deviation e(m) from the reference physiological state as a function of the distances d merger(m) during a seventh step G). As previously discussed, this is possible because the reference physiological state, associated with the reference matrix Xref of components Xref,i, is already known. Regarding the determination of the deviation e(m) from the reference physiological state, it can be further specified that not every deviation is necessarily considered to reflect a change of state. Indeed, a threshold Θ can be defined above which a deviation e(m) from the reference physiological state is considered sufficiently significant to indicate that there is indeed a change in the patient's physiological state. The deviation e(m) that is calculated is that between the merged distances d merger (m) at times t: t+W and the merged distances.

[0063] According to a particular embodiment of the process according to the invention, prior to step G, the merged data is filtered. This filtering step consists of smoothing obtained by a moving average over the last L time segments m of the observation window (1 <L<M) appliquée aux distances résultantes d merger (m). Smoothing aims to further reduce the number of false alarms from the respiratory support device by reducing irregularities typically caused by signal attenuation. Preferably, this smoothing step is implemented by applying the following relationship to d fusion (m): d fusion L m = 1 L ∑ i = 0 L − 1 d fusion m − i

[0064] The value of the fusion distance thus calculated significantly increases the performance of the process. Indeed, the smoothing of the fusion distances of merger (m) provides a better overall classification by reducing their variability, and this for all M time segments.

[0065] Advantageously, the last L segments are chosen such that the moving average is calculated taking into account signals measured only with a delay of between 10 seconds and 1 minute, preferably between 30 seconds and 1 minute, relative to the start of the observation window. By thus lengthening the time window over which the windows are smoothed, particularly to approximately 1 minute, it is possible to considerably improve the performance of the detection method according to this particular embodiment of the invention, as will be seen in more detail below.

[0066] It should be noted that the detection method according to the invention can be implemented using any other data processing method than that described above in relation to steps B) to F). Indeed, in the method according to the invention, what is important is to fuse the data from measurements taken from the different types of signals. Furthermore, if, according to a preferred embodiment, this fusion is implemented by performing a weighted sum of the data obtained at the end of steps B) to E) in order to obtain resulting distances d merger (m) ,Another fusion process based on data other than distances could be considered. In other words, the specific implementation previously described with reference to steps B) to F) is proposed solely to facilitate understanding of the invention. One could consider merging the data from the different measurements using methods known to those skilled in the art, such as methods based on Bayesian probabilities, Dempster-Shafer belief theory, transferable belief models, Dubois and Prade possibility theory, etc. Practical application of the method for detecting a change in a patient's physiological state

[0067] In this embodiment, the method for detecting a change in physiological state according to the invention is implemented for two subjects, SU1 and SU2, based on measurements of electroencephalographic S1 signals and measurements of a respiratory S2 signal for each subject. In this embodiment, the procedure was carried out as described previously, implementing successive steps A to G according to a particular embodiment of the method according to the invention. figures 2a And 2b illustrate the performance obtained with the detection method according to the invention for a SU1 subject ( figure 2a ) and a SU2 topic ( figure 2b ).

[0068] On each of the figures 2a , 2b The first figure illustrates Riemannian distances of 1,r estimated from the S1 electroencephalographic signals alone, while the second figure illustrates the statistical distances of 2 estimated from the respiratory S2 signal alone. In other words, these first two figures represent the performance associated with an independent detector based on the analysis of a single type of signal from among the aforementioned signal types. On each curve in the figures, a first segment m1 and a second segment m2 are distinguished, in which the Riemannian and statistical distances (corresponding to the two physiological states of a patient) have been calculated. On each of the figures 2a , 2b The third figure illustrates the fusion distances obtained after performing the fusion of log d 1,r and log d 2,s according to formula (4) above. We opted for the 90% quantile – ninth decile – to define the precision. Here, the weighting coefficients are identical λ 1 = λ 2 = 0.5.

[0069] Each figure (subfigure) shows a value corresponding to the area under the curve AUC (Area Under the Curve(in English) which allows monitoring of detection performance. For subject SU1, the AUC values ​​associated with distances calculated from the S1 electroencephalographic signals alone, from the S2 respiratory signals alone, and the fusion distances resulting from the fusion of these two types of data are 0.85, 0.74, and 0.91, respectively. The net increase in AUC after data fusion corresponds to a net improvement in performance. For subject SU2, the AUC values ​​associated with distances calculated from the S1 electroencephalographic signals alone, from the S2 respiratory signals alone, and the fusion distances resulting from the fusion of these two types of data are 0.91, 0.72, and 0.91, respectively.In this case, even if the increase in AUC is not as significant as what was seen for subject SU1, there is an improvement in the quality of the merged data compared to the unfused data.

[0070] On each of the figures 2a , 2b The fourth figure illustrates the operating characteristics of the receiver, denoted ROC. (Receiver Operating Characteristic(in English) for each subject. Each figure shows three curves corresponding, from lightest to darkest, to the ROC curve associated with the respiratory S2 signal alone, the ROC curve associated with the electroencephalographic S1 signals alone, and the ROC curve associated with the fusion distances resulting from the fusion of data from these two types of measurements. It can be observed that the false alarm rate increases less rapidly with a detector using the fused fusion distances compared to what is measured by independent detectors, i.e., detectors using the electroencephalographic S1 signals alone or the respiratory S2 signal alone. It can be noted that for subject SU2, the independent detector using the electroencephalographic S1 signals alone nevertheless exhibits comparable performance to that of the detector using the fused data.

[0071] As mentioned in the general embodiment, it is also possible to improve detector performance by filtering the data. While in the method according to the invention this preferred filtering step is applied to fused data, it can also be applied to unfused data. The inventors implemented this smoothing on both fused and unfused data to highlight the performance of the fusion. This is illustrated in the figure 3 .

[0072] An improvement in the performance of each detector can already be observed by smoothing the data obtained after a 15-second delay. Indeed, as can be seen in Figures 3a, 3b, and 3c, respectively, the AUC values ​​associated with each detector increase as this delay increases. While a performance improvement is clearly visible for detectors using only the respiratory S2 signal and only the electroencephalographic S1 signals, it is even more pronounced for the detector using the fused data. The results are remarkable, as shown in Figure 3c. Description of a general embodiment of a patient monitoring device according to the invention

[0073] It should be noted first of all that by patient monitoring device we mean any device as described below as long as it allows the measurement of at least two physiological parameters of a patient, that is to say at least two functions and / or reactions of the patient.

[0074] The invention further relates to a patient monitoring device enabling the implementation of a detection method as previously described. The monitoring device is advantageously equipped with a processor that conventionally performs tasks, including data processing, and communicates with other equipment to which it is connected.

[0075] To implement the method according to the invention, the patient monitoring device includes means for measuring electroencephalographic signals (EEGs). These means typically include a headset equipped with n electrodes appropriately positioned on the patient's head. These means may also include any other equipment enabling electroencephalographic measurements. The headset is connected to the monitoring device and transmits all the data measured by the electrodes to it. The monitoring device is capable of receiving and processing the information received from the headset and, when necessary, adjusting the ventilatory support provided to the patient.

[0076] The monitoring device also includes means for measuring cardiac activity and / or respiratory activity and / or muscular activity and / or movement. In this regard, the monitoring device is linked and / or connected to sensors and other diagnostic tools commonly used to measure the relevant physiological parameters. These sensors or other tools may include cervical electrodes or accelerometers, depending on the type of physiological parameter being observed.

[0077] As previously mentioned, the monitoring device is capable of performing data processing tasks. It includes real-time signal processing means, notably means for determining X-ray components. i,refof the reference matrix, means of calculating the deviations e(m) from the reference situation, and means of merging the data. In any event, it is capable of implementing the process that is the subject of the invention.

[0078] The invention further relates to a medical assistance device equipped with a monitoring device as previously described.

Claims

1. A method for detecting a change in the physiological condition of a patient relative to a reference physiological condition associated with a reference matrix Xref with components Xref,i (i = [1...N+1], N integer) and with a reference period W0, the method being a processor-implemented method and implementing the following steps in a loop: A) in M time segments m of an observation window, performing measurements of electroencephalographic signals S1 from the patient along n paths and at p times to generate M measurement matrices X1,m (m ε [1....M], M integer) each comprising n*p samples, and simultaneously performing measurements of at least N other types of physiological signals SN+1 (N ≥ 1) from the patient, different from the electroencephalographic signals S1, to generate N other measurement matrices Xi = 2..N+1,m, C) for each time segment m of the reference period W0, determining distances di(m) between each measured signal Xi,m and the component Xref,i of the reference period, D) transforming the distances di(m) calculated in step C) using a log di(m) function, E) choosing the precision of the values obtained at the end of step D) by selecting a reference quantile u0 from all the values of the reference period W0 and associating with each distance di(m) a scalar variable pi=1,...,N+1 (0<pi=1,...,N+1<1) depending on the reference quantile u0, F) merging the data obtained in step E) by performing a weighted sum of the variables pi=1,...,N+1 to obtain the resulting distances dfusion(m), G) determining a deviation e(m) from the physiological reference condition as a function of the distances dfusion(m).

2. The method according to claim 1, wherein in a step B) the electroencephalographic signals S1 are centred and filtered in Q predetermined frequency bands to obtain MxQ filtered measurements matrices X1,m,q (q ε [1...Q]) and MxQ normalised spatial covariance matrices are determined by the formula: C m , q = X 1 , m , q X 1 , m , q T trace X 1 , m , q X 1 , m , q T and in which, in step C), Riemannian distances di,r(m) are determined between each normalised spatial covariance matrix Cm,q and the components Xref,i of the reference matrix associated with the electroencephalographic measurements.

3. The method according to any one of claims 1 to 2, wherein, during step C), statistical distances di,s(m) are determined between one or more measured signals Xi,m other than the electroencephalographic signals, and the component or components Xref,i of the reference period associated with it.

4. The method according to any one of the preceding claims, wherein dfusion(m) is calculated according to the following relationship: d fusion m = ∑ i λ i p i = 1 , … , N + 1 m , with p i = 1 , … , N + 1 m = 1 1 + e u 0 − log d i = 1 , … , N + 1 m where λi is the weight assigned to each variable pi=1,...,N+1, the λi verifying ∑iλi = 1, u0 is the reference quantile, d1,q(m) are the Riemannian distances and di>1(m) are the statistical distances.

5. The method according to claim 4, wherein the weights λi are selected according to the quality of the signal Si with which they are associated or according to the importance of the signal Si with which they are associated for determining a change in physiological condition in the patient.

6. The method according to any one of the preceding claims, wherein a step of filtering the resulting distances dfusion(m) is implemented prior to step G).

7. The method according to claim 6, wherein the filtering consists of a smoothing obtained by a moving average over the last L segments of the observation window (1<L<M) applied to the resulting distances dfusion(m), said distance dfusion(m) then being calculated according to the following relationship: d fusion L m = 1 L ∑ i = 0 L − 1 d fusion m − i 8. The method according to claim 7, wherein the last L segments are chosen so that the moving average is calculated taking account of the measured signals only with a delay of between 10 seconds and 1 minute, preferably between 30 seconds and 1 minute, relative to the start of the observation window.

9. The method according to any one of the preceding claims, wherein said other physiological signals Si are cardiac activity and / or respiratory activity and / or muscular activity and / or a movement-related signal.

10. The method according to any one of the preceding claims, further comprising a step during which each of the deviations e(m) is compared with a determined threshold Θ.

11. A device for monitoring a patient for the implementation of a method according to any one of the preceding claims, comprising: - electroencephalographic signal measuring means for measuring the patient's electroencephalographic signals S1, - means for measuring cardiac activity and / or respiratory activity and / or muscular activity and / or movement, - real-time signal processing means, said means comprising at least means for determining the components Xi,ref of the reference period, means for calculating the deviations e(m) from the reference situation and means for merging the data, - a processor able to implement the method for detecting a change in the physiological condition of a patient relative to a reference physiological condition according to any one of claims 1 to 10.

12. A medical assistance device comprising a monitoring device according to claim 11.