Method for determining the state of an individual in relation to reference states

Abstract

Method for processing physiological data of an individual, each physiological data resulting from measurements carried out on the individual, the method comprising the establishment of a score of the individual, of type z score, and a projection of the z-score in a space so as to determine a state of the individual.

Description

Title of the invention: Method for determining the state of an individual in relation to reference states Technical field

[0001] The technical field of the invention is the processing of data, for the purposes of assisting in the diagnosis or prognosis of a pathology of an individual. PREVIOUS ART

[0002] Biological or medical examinations carried out on patients are intended for a practitioner to assist in diagnosis. A common practice consists of comparing physiological characteristics, determined on a patient, with so-called reference physiological characteristics, established on a reference population considered to be healthy.

[0003] The digitalization of examinations allows the creation of large databases for different types of examinations, whether they are biological analysis results or data obtained by imaging modalities or other types of measurements carried out on a patient for diagnostic purposes. These may be, for example, electrophysiological measurements such as ECG (Electrocardiogram) or EEG (Electroencephalogram).

[0004] The constitution of databases makes it possible to determine, with greater statistical precision, the reference physiological characteristics. It also allows for better individualization of the reference values: contextual data (age, weight, height, sex) or physiological data resulting from other types of examination can be taken into account to refine the reference physiological characteristics.

[0005] However, the databases are generally large. Dimensionality reduction may be relevant, as described in the publication Attyé et al “Tractleam: a geodesy learning framework for quantitative brain bundles”. In this publication, we describe the use of a dimensionality reduction algorithm by manifold learning, usually referred to as “manifold learning”, for the analysis of brain structures from characteristics resulting from nuclear magnetic resonance imaging (MRI).

[0006] The inventors propose a method for processing physiological data, exploiting dimensionality reduction for diagnostic assistance purposes. Statement of the invention

[0007] A first object of the invention is a method for processing physiological data of an individual, each physiological data resulting from measurements carried out on the individual, the process comprising the following steps: - a) based on the physiological data of the individual, assignment of coordinates, called origin coordinates, in an origin space, to the individual, each origin coordinate being assigned a rank, the origin coordinates forming a vector of characteristics; - b) projecting the feature vector into a first latent space to form a latent vector; - c) applying a regression function to the latent vector resulting from b), so as to estimate an original coordinate, in the original space, for each rank; - d) for at least one rank, comparison of the original coordinates of the individual resulting from c) and a) and determination of a score for the individual based on the comparison; - e) formation of a score vector from scores determined for different physiological characteristics;

[0008] the method being characterized in that it comprises - f) projection of the score vector into a second latent space, so as to obtain a projected score vector, the second latent space comprising at least one state group, corresponding to coordinates, in the latent space, associated with a state of the individual; - g) determining a position of the projected score vector relative to a state group or each state group;

[0009] steps a) to f) being implemented by computer.

[0010] The state of the individual can be: - at least one pathological condition or one healthy condition; - and / or at least two different pathological states, corresponding respectively to at least two different pathologies; - and / or the same two different states of the same pathology.

[0011] According to one possibility: - the first latent space is determined from physiological data established on reference individuals from a reference population; - the second latent space is determined from physiological data established on individuals of different states.

[0012] Step f) may include a projection of the individual's score and at least one contextual characteristic of the individual. The contextual characteristic may include an age of the individual and / or a sex of the individual, or even morphological data of the individual, such as a height or a weight.

[0013] The method may comprise, prior to step a), obtaining the data physiological of the individual using at least one sensor.

[0014] The method may comprise: - formation of a two-dimensional or three-dimensional image of the individual; - extraction of physiological data from each image of the individual.

[0015] The image may result from a medical imaging modality such as MRI, CT scan, scintigraphy, infrared spectroscopy, Raman spectroscopy.

[0016] The physiological data may result from an analysis modality such as: electrocardiogram, electromyogram, electroencephalogram, magnetoencephalography, oximetry, analysis of the composition of body gas.

[0017] The method may comprise an analysis of a bodily fluid of the user, previously taken, the physiological data of the individual resulting from the analysis of the bodily fluid.

[0018] Another object of the invention is a system for processing physiological data of an individual, the system comprising: - a sensor, or a set of sensors, configured to measure physiological data of the individual; - a processing unit, configured to receive the physiological data measured by the sensor or set of sensors, and to implement steps a) to f) of a method according to the first subject of the invention.

[0019] A second object of the invention is a system for determining a state of an individual, the system comprising: - a sensor, or a set of sensors, configured to measure physiological data of the individual; - a processing unit, configured to receive the physiological data measured by the sensor or set of sensors, and to implement steps a) to f) or a) to g) of a method according to the subject of the invention.

[0020] Another object of the invention is a support, connectable to a computer, comprising instructions for implementing steps a) to f) of a method according to the first object of the invention from physiological data measured on an individual.

[0021] The invention will be better understood upon reading the description of the exemplary embodiments presented in the remainder of the description, in conjunction with the figures listed below. FIGURES

[0022] [Fig.l] shows an example of a system suitable for implementing the invention.

[0023] [Fig.2] shows schematically the main steps of implementing a method according to the invention.

[0024] [Fig.3] schematizes a projection into a latent space

[0025] [Fig.4A] shows groups of states, formed by latent vectors of different individuals whose states are known. The latent vectors are defined in a first latent space.

[0026] [Fig.4B] shows state groups, formed by projected score vectors of different individuals whose states are known. The projected score vectors are defined in a second latent space.

[0027] [Fig.4C] is a figure similar to [Fig.4B]. In [Fig.4C], contextual data, in this case the user's age, has been taken into account in the score vector before its projection into the second latent space.

[0028] [Fig.5] shows groups of states of individuals with the same pathology and control individuals. The pathological individuals are divided into two states, depending on the sensitivity of the pathology to a treatment. PRESENTATION OF SPECIAL METHODS OF IMPLEMENTATION

[0029] [Fig. 1] represents an example of a system allowing an implementation of the invention. The system 1 comprises a system 10 for acquiring physiological data from an individual. In this example, the acquisition system is an MRI remnography system. The individual is a human or animal being. By physiological data, we mean a value of a physiological parameter of the individual, likely to vary depending on the state of health of the individual. It may be data resulting from an image acquisition system, or from a system for acquiring electrophysiological measurements (for example ECG - Electrocardiogram, EEG - Electroencephalogram, EMG - Electromyogram, MEG - Magnetoencephalography, oximetry or any other analysis of a body gas, in particular a respiratory gas). It may also be data measured on a sample previously taken from the patient.The sample may be, for example, a bodily fluid, such as blood or urine. The data may be a concentration of a cell or protein or other molecule of biological interest in the bodily fluid. When the measured data result from an analysis of a bodily fluid, the physiological data acquisition system may be a biological analysis automaton.

[0030] The system 1 comprises a processing unit 12, programmed to receive the data measured by the acquisition system and implement the processing steps described below. The processing unit may be a computer or a remote cloud-type computing architecture.

[0031] An objective of the invention is to enable assistance in determining a possible pathology from the measured physiological data.

[0032] The physiological data of the individual, measured by the acquisition system 10 and processed by the processing unit 12, are preferably numerous. The number of physiological data is preferably greater than 5 or even 10. It is typically several tens. As in the publication cited in the prior art, the method implements a variety learning method using linear or non-linear dimensionality reduction methods. Different data dimensionality reduction methods are available. Generally speaking, a dimensionality reduction method makes it possible to represent original coordinates, in an original space (or real space) R of dimension / , into coordinates in an arrival space E of dimension H, with I > H. The arrival space is usually referred to as the "latent space". As previously indicated, the dimension of the original space R may be a few tens. The dimension of the arrival space E is preferably less than 10.The original coordinates, or features, are formed from the individual's physiological data. The original coordinates form a feature vector.

[0033] The dimension of the arrival space depends on the number of physiological characteristics and their complexity. Preferably, it is less than 10 so as to facilitate the performance of calculations.

[0034] The passage from the original space R to the latent space E is carried out by a projection function f applied to the vector formed from the physiological characteristics, called the characteristic vector. The passage from the latent space E to the original space R is carried out by a backprojection function applied to a vector formed by the coordinates in the latent space.

[0035] The latent space E, as well as the projection and backprojection functions are defined during a learning phase, during which a set of reference physiological characteristic vectors is available. Each reference physiological characteristic vector comprises data measured on reference individuals forming a reference population. The objective of dimensionality reduction is to obtain a representation of the reference physiological characteristics in the latent space, the latter being of restricted dimension compared to the original space.

[0036] Several dimensionality reduction methods are known to those skilled in the art. For example, principal component analysis (PCA) is an unsupervised and linear dimensionality reduction method. The basis of the latent space is formed by eigenvectors of the covariance matrix of the feature vectors. Nonlinear methods can be implemented, for example a multidimensional scaling (MDS) type method, preserving the distance values ​​between all pairs of data in the original space. An example of an MDS method is the Sammon method.

[0037] Among the non-linear methods, we can also mention the Isomap (Isometric feature mapping) method, which allows defining a transformation that preserves the geodesic distance between the data. Another type of method is the UMAP (Uniform Manifold Approximation and Projection) method.

[0038] [Fig.2] shows the main steps of a method according to the invention. We are first interested in the learning phase, allowing the definition of latent spaces and projection and back-projection functions.

[0039] Steps 200 to 260 describe an implementation of a method for processing characteristics of a test individual, for which it is desired to establish diagnostic assistance. By diagnostic assistance, we mean a provision of information relating to the state of health of the individual, forming an aid to a practitioner to establish a diagnosis. Steps 100 to 190 are steps for learning the method.

[0040] Step 100: Formation of reference vectors.

[0041] During this step, a set of physiological data measured on reference individuals wf j forming a reference population is available. The reference population is formed of individuals whose state is considered to be known, and preferably healthy.

[0042] For each reference individual, from the measured physiological data, characteristics XiJ are established forming a vector of characteristics Xj defined in the original space. The index i corresponds to a rank assigned to each physiological data, with \

[0043] In this example, each characteristic is a measured physiological data centered and reduced:

[0044] r _ ( । \ (fi V 17

[0045] where is an average of the values ​​m'< for all 7 reference individuals andC is the standard deviation of the characteristics Xi-i of the same rank i. ​

[0046] The characteristics of each reference individual form a reference vector of dimension I, assigned to the individual ref.. Each characteristic x^ corresponds to an origin coordinate of the reference individual ref. in the origin space R.

[0047] For example, each measured value mi~j can be an analysis of a parameter resulting from a blood analysis or an image. For each reference individual ., a reference vector Xj is established comprising the measured values, for the reference individual, centered and reduced taking into account the entire reference population.

[0048] Step 110: Dimensionality reduction

[0049] During this step, a dimensionality reduction algorithm is applied to all or part of the reference characteristics of the reference vectors %j.

[0050] The dimensionality reduction algorithm defines a first linear or non-linear projection function f}. This is, for example, a UMAP type algorithm.

[0051] The projection function allows the vector X / to be projected into a first latent space E} of dimension less than the dimension of the original space R. In the first latent space E^ we obtain a projected vector V ■ = f whose dimension H । is less than the dimension of the vector A / . The first latent space Ex is usually called a “variety” by those skilled in the art.

[0052] The dimension of the projected vector V j is usually between 1 and 10. Each term vh,j, of the projected vector is a coordinate of the reference individual ref. in the latent space. The index h corresponds to a rank assigned to each term vhj. h is an integer between 1 and H being the dimension of the first latent space E{. The dimension is usually between 1 and 10.

[0053] Step 120: Formation of the latent vector

[0054] The terms vh,j of the projected vector Vj can form a latent vector Yj of dimension H],

[0055] According to one possibility, for each rank h, the terms of the projected vector Vj can be centered and reduced, so as to obtain centered coordinates y\j reduced in the latent space:

[0056] _ AÇi / ? j

[0057] where vh is an average of the terms for all J reference individuals for the same rank A, and is the standard deviation of the terms vhj of the same rank h for all J reference individuals.

[0058]

[0059]

[0060] For each reference individual, the set of reduced centered coordinates forms a latent vector Yj of dimension H Thus, the latent vector Y / comprises either the ^hj terms of the projected vector V j, in which case Y ; = V j, or the reduced centered terms as described in connection with (2). A backprojection (or regression) function is used to estimate an original vector v. such that y . _ yj has the same dimension as Aj. Each term of the vector % is an estimated characteristic Xjj in the original space R, from the latent vector Y,. The backprojection function can be established from an es Nadaraya-Watson kernel estimator. Such an estimator is commonly used to establish nonparametric regression models. It is constructed from a kernel function for example Gaussian, and a window of size h. When the projected vector V j is not centered and reduced, Y j— V j and y. _ y ) — fU y

[0061]

[0062]

[0063]

[0064]

[0065]

[0066]

[0067]

[0068]

[0069]

[0070]

[0071]

[0072]

[0073]

[0074] According to one possibility, described later, a regression function r1 is de- U terminated for each rank i. Thus, from the latent vector Y we estimate a characteristic Xq of rank (in the original space R such that: F1 is a regression function determined for the characteristic of rank i. According to ' U in this variant, two regression functions, applied to the same latent vector, and allowing the estimation of two different original coordinates, are different. Step 130: Backprojection and estimation of the feature vector into the original space R. Applying the backprojection function to the latent vector Y makes it possible to obtain an estimate of a vector x, defined in the original space R, as described in step 110. y. is such that: X can be considered as a digital twin of Xj sj is a deviation vector, of the same dimension as X- and F corresponds to a residual of the regression to the model. The more the reference individual ref. is representative of the reference population, the lower the norm of fjk. The deviation vector eJ includes residuals eF respectively associated with each characteristic of rank i: The distribution of values ​​varies depending on the reference individual ref. and the rank of characteristic i. The higher the dimension H {, the lower the average of the S'-J residues of each individual. It is considered that beyond a dimension of 3 or 4, the gain in terms of average residue decreases or no longer evolves significantly. Each residue can be used to determine a z-score for individual i, relative to characteristic j. The term "Z-score" is a common term for those skilled in the art. The determination of the first latent space and the first projection function are carried out in such a way as to minimize the residuals for the reference population. The first latent space E^ and the first projection function. / Â are established by putting implement a dimensionality reduction technique. This involves preserving or maximizing variability between individuals, or allowing separation of individuals by class, while reducing the dimension

[0075] Steps 100 to 130 correspond to a first learning, during which the first latent space E^ is defined, as well as the first projection function / 1 and the backprojection function

[0076] Steps 140 to 180 describe a second learning process, in which the first learning process is used to define a second latent space different from the first latent space Eu in which at least one state group Gs is defined. The or each state group is associated with a state, chosen from a healthy state or a pathological state, the pathology being known. Each state group Gs is a set of coordinates, in the second latent space, which correspond to the same state s.

[0077] More generally, the state of the individual may correspond to: a type of pathology, a characteristic of a pathology (for example Alzheimer's disease with atrophied or non-atrophied hippocampus), or a state of evolution of a pathology or a severity of a pathology or an inflammatory state. The pathological state may also represent a prediction of evolution of a pathology, or a typology of a pathology, for example a sensitivity of a pathology to a treatment.

[0078] Step 140: Formation of a feature vector

[0079] The implementation of step 140 assumes the consideration of a learning population, comprising individuals k considered as having a known state. The state is either a known pathological state or a healthy state. Preferably, the learning population is formed of individuals in a state chosen from at least two different pathological states: healthy state and pathological state, or at least two different pathological states, that is to say: two different pathologies, or two different typologies of the same pathology (different clinical scores for the same pathology, different levels of severity or progress of the same pathology, sensitivity to a different treatment of the same pathology). For each individual of the learning population, there is a vector of characteristics Xks as defined in step 100.Each feature is centered reduced according to (1) if this option was chosen during step 100, to define the first latent space.

[0080] Step 150: Projection into the first latent space E} of each vector

[0081] In this step, the respective feature vectors of each individual k having a state are projected into the first latent space E^ so as to form a latent vector hk.s, as described in connection with step 120. For each individual k, the first projection function f is implemented so as to obtain a projected vector y, . — f ( ) • The vector Lkj is obtained from Xks as pre- previously described.

[0082] Step 160: rear projection

[0083] During this step, the regression function is implemented so as to estimate, from each latent vector Y, the original vector y such that [ yy The regression function is the one determined in step 120. KyS J | y nS J Step 170

[0084] During this step, we establish, for each characteristic z of each individual k, presenting the state, a Z-score Hkj: for each characteristic i.

[0085] such that:

[0086] _ _ LlùLl — (6)

[0087] Generally speaking, the z-score involves a comparison between and s. In the example of formula (6), the comparison is a subtraction normalized by the standard deviation which corresponds to the standard deviation of the values ​​of the residuals determined for the characteristic i for each reference individual.

[0088] Step 180: formation of a score vector.

[0089] During this step, from the z-scores calculated during step 170, a vector of scores is established for each individual k considered to be in state v. The dimension of the vector of scores corresponds to the number of z-scores that are considered relevant for the following step.

[0090] The score vector may include terms representative of contextual characteristics, for example the weight or height or age or gender of the user. Thus, the dimension of the score vector is 1,1' which may be equal to or different from I.

[0091] Step 190: formation of a second latent space and determination of a second projection function

[0092] During this step, a second latent space E^ is determined as well as a second projection function allowing a projection of a score vector, defined in the original space R, into the second latent space. The dimension of the second latent space E2 can be equal to or different from the dimension H ] of the first latent space Ev. The second projection function makes it possible to form a projected score vector Wk,= f2(Zk^. (7).

[0093] The second latent space E2 is such that the projected score vectors of the individuals k presenting the same state J are grouped, in the second latent space, so as to form a state group GsDc same as for the first latent space, the second latent space is dimensioned to capture the variability of the learning population in a space of limited dimension, preferably less than 5, or even less than or equal to 3. The second latent space is then defined so as to allow, in a space of restricted dimension, a better possible separation between individuals of different states.

[0094] Steps 150 to 190 are carried out using physiological data measured with individuals presenting at least two different and known states: healthy state / pathological state (presence of a pathology), two different pathological states, as previously described, two different types of the same pathology, different clinical scores for the same pathology.

[0095] Steps 140 to 190 may be performed using physiological data measured with training individuals exhibiting different pathologies. In this case, the second latent space is defined so as to separate each pathology as best as possible. The second latent space is then defined so as to optimize a separation between healthy individuals and individuals exhibiting each pathology.

[0096] More generally, during steps 140 to 190, the learning population is formed of individuals whose state, pathological and / or healthy, is known. The second latent space is determined so as to separate the individuals as best as possible according to their state. In the second latent space, the coordinates corresponding to each state are respectively grouped together to form state groups Gs.

[0097] Steps 200 to 250 correspond to an implementation of the invention on a test individual. The test individual is assigned a vector of physiological data mifest Ç)n understands that for each rank, the physiological data of the test individual are of the same nature as the physiological data established for each reference individual or each individual of the learning population.

[0098] Step 200: From the measured physiological data, characteristics are established forming a characteristic vector Xtest defined in the original space R. In this example, each characteristic is a measured physiological data centered and reduced:

[0099] r _ / in )

[0100] and were defined in step 100.

[0101] Step 210: projection of the feature vector into the first latent space Ev

[0102] During this step, the vector X!est is projected into the first latent space defined in the first learning step (steps 100 to 130), so as to obtain a projected vector = f (Xtei!f ) • The projection is carried out using the first projection function f { previously established following the first learning.

[0103] The vector Vtest has coordinates with 1 < h < Hx. Hy is the dimension of the first latent space E^.

[0104] In [Fig.3], we have schematized the main vectors implemented in the steps 200 to 260.

[0105] Step 220: Obtaining latent coordinates

[0106] Optionally, the projected vector Viesr can be centered and reduced, so as to obtain reduced centered y^est coordinates in the first latent space:

[0107] = 22^ hdest. (11)

[0108] where / is an average of the terms vhj for the set of J reference individuals for the same rank A, and is the standard deviation of the terms vh,j of the same rank h for the set of J reference individuals. / is determined during the first phase previously described learning. See expression (2) of step 120.

[0109] When the projected vector Vtest is not centered and reduced, each latent coordinate yhjest is such that Yh / est ~ viuest, Thus, Ytest = Vtest

[0110] The coordinates Y^est form a vector Ytest of dimension H j in the first latent space E^. [YES] Step 230: rear projection

[0112] During step 230, the regression function corresponding to the latent space El is applied to the vector Y to estimate a vector x such that X( t = ( Y tt ) • Each term of the vector Xtest is an estimated characteristic Xi!est ,in the original space R, from the vector Y tes( of the coordinates in the first latent space E^.

[0113] Step 240: calculation of a score

[0114] During this step, the z-score zUesP of the test individual is determined, and this for each characteristic of rank 1 ' tJesE / iq \ — X 1 3 /

[0115] was previously defined.

[0116] According to one possibility, the z-score E.test is determined according to the expression:

[0117] (14) ^'utest (r'e-

[0118] <7 / is determined according to a procedure called “leave one out”, according to which the Iterative steps 110 and 140 are performed by successively removing a reference individual of rank / . This makes it possible to form J first latent spaces / / (. Each first latent space E^j is defined from the reference population, without taking into account the reference individual of rank j'.

[0119] Subsequently, steps 200 to 230 are implemented using the individual of rank / as a test individual. Thus, for each individual of rank j \ a residue is obtained such that = x,- (5').

[0120] vEl then corresponds to the standard deviation of the residues sij corresponding respectively to the different individuals of rank j 'considered.

[0121] The reference individuals rank j' can be chosen from among the reference individuals closest to the test individual, for example the 30 closest.

[0122] The different z-scores, respectively established for each characteristic i, allow a formation of a vector of scores Ztor, each term of which is either a z-score or a contextual characteristic, for example chosen from the age and / or the sex of the individual and / or the weight or the height or other morphological criteria, for example body mass index.

[0123] Step 250: projection of the Ztest score vector.

[0124] During this step, the score vector Ztes( is projected into the second latent space E2 so as to obtain a projected score vector = f ^Ztest ) • The projection is performed using the second projection function , / 7 previously established during step 190. Stage 260

[0125] As previously described, the second latent space E2 comprises state groups G s grouping coordinates corresponding respectively to different states of the user: healthy state and / or one or more pathological states. When the coordinates of the projected score vector Wfest are part of a state group G^, an indication is obtained according to which the test individual is likely to be affected by the state5 associated with this group.

[0126] The method comprises a determination of a position of the vector W?eîf relative to each state group Gs. The diagnosis of the individual, depending on the position, is the responsibility of the practitioner. Thus, the invention can be limited to a provision of information allowing the practitioner to make a diagnosis.

[0127] The dimension of the or each state group Gs. corresponds to the dimension EI2 of the second latent space / L-In the examples described below, the second latent space has a dimension equal to 2. Each state group is then formed by a surface or a union of several surfaces. When the dimension of the second latent space is equal to 3, each state group Gs is a volume or a union of volumes.

[0128] The second latent space may comprise several state groups G^, each state group being associated with a different state (for example a different pathology) from that associated with another state group. The second latent space may comprise a state group considered to be representative of a healthy state of the test individual. When the coordinates of the vector are included in the latter, the test individual is considered to be healthy.

[0129] Experimental tests.

[0130] During a first series of tests, the physiological data determined were volumes of brain structures observed on anatomical images

[0131]

[0132]

[0133]

[0134]

[0135]

[0136]

[0137]

[0138]

[0139] obtained by MRI, with Tl weighting. For learning, we took into account: - 425 learning individuals with Alzheimer's disease (AD = Alzheimer's Disease), whose characteristics were extracted from the international cohorts AIBL (Australian Imaging Biomarker and Lifesyle), MIRIAD (Minimal Interval Resonance Imaging in Alzheimer's Disease) and ADNI (Alzheimer's Disease Neuroimaging Initiative); - 296 learning individuals with frontotemporal dementia (FTD), whose characteristics were extracted from the international cohorts NFID (Northern Finland Intellectual Disease) and 4RTNI (R Repeat Tauopathy Neuroimaging Initiative) - 537 learning individuals with Parkinson's disease (PD: Parkinson Disease), whose characteristics were extracted from the international PPMI (Parkinson's Progression Markers Initiative) cohort. The interest of these cohorts is that they were established from several MRI devices, with different Tl sequence settings. We also considered 398 healthy learning individuals (C: control). These were used as reference individuals, to form the first latent space during the first learning, and as learning individuals, to form the second latent space during the second learning. Steps 100 to 200 were implemented taking into account: - for steps 100 to 130: healthy individuals; - for stages 140 to 200: healthy individuals and pathological individuals. For each individual, 196 characteristics from Tl MRI imaging sequences were taken into account. Figure 4A shows the first latent space, of dimensions 2, in which the reduced non-centered latent vectors of individuals have been grouped. Figure 4B shows the second latent space, of dimensions 2, in which the projected score vectors Wk have been grouped <s des individus. In each figure, we have represented state groups Gg, gathering the vectors Fkj (figure 4A) and W (figure 4B) of individuals of the same state \ We observe that in figure 4B, the state groups are better separated than in figure 4A. This shows that if we wish to separate the individuals according to their state, the vector Wk.s is more relevant than the vector Y ks, Figure 4C shows an embodiment in which an age of each individual has been taken into account in the score vector of each individual. Then the score vector Zks has been projected into the second latent space, so as to form a projected score vector Y. It is observed that in Figure 4C, the state groups Gs are better separated than in [Fig.4B]. Thus, taking contextual data into account can improve the separation of individuals according to their condition. In [Fig.4C], each set is delimited by an annotated contour according to the percentage of individuals with the same condition that it delimits: for example, AD80 corresponds to a contour containing 80% of individuals with Alzheimer's disease.

[0140] In a second series of tests, 196 cortical structures were analyzed for control subjects (group C), as well as for subjects suffering from multiple sclerosis (MS), with different levels of effectiveness for a given first-line treatment: MS group 1 corresponds to individuals for whom the first-line treatment works moderately. MS group 2 corresponds to individuals for whom the first-line treatment does not work. In this example, there are 40 individuals in each MS1 and MS2 group and 100 individuals in the control group C.

[0141] [Fig.5] shows a projection, in the second latent space, obtained with all the individuals taken into account: the control populations C, SEP 1 and SEP 2 can be separated. Thus, the implementation of the invention makes it possible to estimate, for an individual presenting a pathology, in this case multiple sclerosis, the type of pathology, as a function of a sensitivity to a treatment.

[0142] The invention makes it possible to form, in the second latent space, an “atlas” of states, in which the projected score vectors are grouped according to a state of a user. The position of the projected score vector forms an assistance to a practitioner to evaluate the state of the user and form a diagnosis.

[0143] A notable advantage of the method is that it allows measured physiological data to be taken into account, by limiting a bias linked to the measuring instrument. The first reduced space makes it possible to capture the effects linked to inter-individual variability as well as to the variability linked to the measuring instrument, in particular when the reference population is sufficiently large: the characteristics of each individual are measured with different devices, which makes it possible to take into account the instrument variability, which corresponds to the variability affecting the measurement parameters of each instrument.

[0144] The projection into the second latent space, from Zscores, makes it possible to limit the influence of instrument variability. This makes it possible to obtain a clearer separation according to the state of each individual: the pathology is thus visualized more specifically and the different pathological states can be better separated.

[0145] The invention may be implemented using physiological data resulting from different types of analysis, for diagnostic assistance purposes.< / s>

Claims

Claims

1. Method for processing physiological data of an individual, each physiological data resulting from measurements carried out on the individual, the method comprising the following steps: - a) depending on the physiological data of the individual ^Gest), assignment of coordinates, called origin coordinates (in an origin space (R), to the individual, each origin coordinate being assigned a rank (i), the origin coordinates forming a vector of characteristics; - b) projecting the feature vector into a first latent space to form a latent vector; - c) application of a regression function ( f1 r1 ) to the latent vector J i ' J li resulting from b), so as to estimate an original coordinate in the original space (R), for each rank (i); - d) for at least one rank ( / ), comparison (&ifest-Xitest) of the original coordinates of the individual resulting from c) and a) and determination of a score of the individual (Zijest) based on the comparison; - e) formation of a vector of scores (Z, test) from scores determined for different physiological characteristics; the method being characterized in that it comprises - f) projection of the score vector (Zîest) into a second latent space, so as to obtain a projected score vector (Wtes[), the second latent space comprising at least one state group (Gy), corresponding to coordinates, in the latent space, associated with a state of the individual; - g) determining a position of the projected score vector relative to a state group or each state group; steps a) to f) being implemented by computer.

2. The method of claim 1, wherein the state of the individual is selected from: at least one pathological condition or one healthy condition; - and / or at least two different pathological states, corresponding respectively to at least two different pathologies; - and / or to the same two different states of the same pathology.

3. Method according to claim 1 or claim 2, in which - the first latent space is determined from physiological data established on reference individuals (J), of a reference population; - the second latent space is determined from physiological data established on individuals (Æj of different states a

4. A method according to any preceding claim, wherein step f) comprises a projection of the individual's score and at least one contextual characteristic of the individual.

5. Method according to claim 4, in which the contextual characteristic comprises an age of the individual and / or a sex of the individual and / or morphological data of the individual.

6. Method according to any one of the preceding claims, comprising, prior to step a), obtaining physiological data from the individual using at least one sensor.

7. Method according to claim 6, comprising - forming a two-dimensional or three-dimensional image of the individual; - extracting physiological data from each image of the individual.

8. Method according to claim 6, in which the image results from a medical imaging modality such as MRI, CT, scintigraphy, infrared spectroscopy, Raman spectroscopy.

9. Method according to claim 6, in which the physiological data result from an analysis modality of the type: electrocardiogram, electromyogram, electroencephalogram, magnetoencephalography, oximetry, analysis of the composition of body gas.

10. A method according to claim 6, comprising an analysis of a fluid body fluid of the user, previously taken, the physiological data of the individual resulting from the analysis of the bodily fluid.

11. A method according to any preceding claim, wherein the latent vector is obtained by projecting the feature vector into a first latent space followed by processing to center and reduce the coordinates of the projected feature vector.

12. A method according to any preceding claim, wherein the latent vector is obtained by projecting the feature vector into a first latent space followed by processing to center and reduce the coordinates of the projected feature vector.

13. System (1) for processing physiological data of an individual, the system comprising: - a sensor (10), or a set of sensors, configured to measure physiological data of the individual; - a processing unit (12), configured to receive the physiological data measured by the sensor or the set of sensors, and to implement steps a) to f) of a method according to any one of the preceding claims.

14. Support, connectable to a computer, comprising instructions for implementing steps a) to f) of a method according to any one of claims 1 to 12 from physiological data measured on an individual.

Images