Method and system for monitoring cognitive state of an individual

Abstract

The Abstract contains an unclear phrase in the description of the input measured data, which hinders a clear understanding of the invention's fundamental inputs. Specifically, in the sentence describing the processing of input measured data, the phrase: "...sensing data of at least one sensing type corresponding to a sending signal being originated on at least one body part of the individual..." is ambiguous. The ambiguity lies in the interpretation of "corresponding to a sending signal." It is unclear what "sending signal" refers to in this context. For improved clarity, the applicant is invited to revise this phrasing. For example, a clearer phrasing might be: "...processing input measured data, indicative of time variation of sensed data of at least one sensing type being originated on at least one body part...". This proposed rephrasing removes the ambiguous "corresponding to a sending signal" while directly linking the sensed data to its origin on the body part.

Description

[0001] METHOD AND SYSTEM

[0002] FOR MONITORING COGNITIVE STATE OF AN INDIVIDUAL

[0003] TECHNOLOGICAL FIELD AND BACKGROUND

[0004] The present disclosure is generally in the field of monitoring the state / condition of an individual during certain activity and relates to a system and method for monitoring cognitive state of the individual in relation to the specific task performance, in particular useful for drivers of vehicles.

[0005] Various techniques have been developed aimed at monitoring individual's state / condition based on brain commands' detection, using EEG signals and / or movements. For example, such techniques are described in US10,413,246; US11,141,113; WO22123568; WO23242842, all being assigned to the assignee of the present application. According to such technique, motion-related data, collected by sensors from one or more parts of the individual's body during certain activity (e.g., task performance), is analyzed to determine a change in the cognitive state from the normal state. Some of these techniques also enable to identify the source of such change.

[0006] GENERAL DESCRIPTION

[0007] There is a need in the art for a novel approach for monitoring / detection of the cognitive state of an individual enabling higher accuracy and confidence of the cognitive state detection. For example, the cognitive state may include but is not limited to drunkenness, fatigue, and error detection.

[0008] The present disclosure provides a novel technique for such high-confidence cognitive state detection utilizing processing and analysis of raw measured data, indicative of sensing data of at least one sensing type corresponding to a sensing signal being originated on at least one body part of the individual during a sensing time period under predetermined measuring conditions. The processing and analyzing of the raw measured data is based on the novel technique of the present disclosure according to which at least one novel model is applied to at least one selected portion of the raw measured data, where the selected portion of the measured data corresponds to a characteristic pattern of the sensing signal sensed over at least one characteristic time segment of the sensing time period. The model is configured to describe a relation between a change in the cognitive state of the individual and a change in the measured data within said at least one characteristic time segment.

[0009] The characteristic time segment is the time segment identified (during the model training procedure) as that of the high-confidence portion of the measured data. This is identified as best-fit conditions (optimal merit functions) for two or more different models, each describing a change in the cognitive state of an individual affecting a change in measured data in relation to the similar activity / task, while differing from one another in sensing types and / or ground truth data types.

[0010] In some embodiments, the sensing data comprises motion data indicative of motion performed by the at least one body part or motion performed by an element / device operated by the at least one body part of the individual. Alternatively or additionally, the sensing data comprises data indicative of autonomous nervous system (ANS) activity.

[0011] In some embodiments, the ground truth data type describes optimal correlation between objective data (e.g., blood alcohol concentration level for drunkenness, or fatigue level according to Karolinska Sleepiness Scale (KSS)), corresponding to the change of the cognitive state of an individual, and the sensing data of the certain sensing type being collected during a given task being performed by the individual.

[0012] Alternatively or additionally, the ground truth data type describes correlation between error-related ground truth (determined based on EEG measurements and / or ANS measurements) corresponding to the change of the cognitive state of an individual, and the sensing data of the certain sensing type being collected during a given task being performed by the individual.

[0013] The above-described two or more models provide determination of the matching measured data portions corresponding to the best-conditions of different-type measurements (differing in the sensing data type and / or ground truth data type) which are independently affected by the same change in the cognitive state of the individual. This enables a desirably high level of confidence in the cognitive state determination. It should be noted that determination of the cognitive state of an individual with high level of confidence is very important in various applications (i.e., various task performances) requiring decision making based on the determined cognitive state, e.g., while monitoring the vehicle's driver’s condition.

[0014] In some embodiments, during the models' training stage, the cognitive states of individuals are intentionally and controllably manipulated while simultaneously recording time evolution of measured data of different types: first-type measured data based on movement kinematics (motion data) and second-type measured data based on the autonomous nervous system (ANS) measures of the individual. The ANS measures may include one or more of the following: heart rate, galvanic skin response, pupil dilation, activity over the corrugator supercilia.

[0015] Thus, first and second measured data simultaneously recorded over time from an individual are analyzed to identify at least one pair of first and second time segments (of the first and second measured data types, respectively) which are at least partially overlapping, and which correspond to the measured data portions of the best fit with corresponding first and second theoretical data (ground truth). Based on this training procedure, a novel high-confidence machine learning model(s) is / are created describing the relation between the cognitive state of the individual and the respective measured data over the respective time segment(s).

[0016] The high-confidence machine learning model(s) (i.e., motion-related model and / or ANS-relating model) can then be used for model-based processing of real-time measured data of the respective type(s) and providing reliable information about possibly impaired cognitive state of the individual.

[0017] It should be noted that movement kinematics are task dependent, and the same is true for ANS measures. Accordingly, the task performed by individuals during the models' training is to be identical to the ecological (real-time) task during which the individual's cognitive state is to be monitored / controlled. For example, if the model is configured to operate for driving scenarios, the training data is collected from drivers in simulation or real driving. Thus, according to one broad aspect of the present disclosure, there is provided a system for monitoring a cognitive state of an individual during a certain task performance, the system comprising: a control system configured as a computer system comprising data input and output utilities, a memory, and a data processor and analyzer, the control system being configured and operable for processing input measured data, indicative of time variation of sensing data of at least one sensing type corresponding to a sending signal being originated on at least one body part of the individual over time under predetermined measuring conditions, to determine the cognitive state of the individual, wherein said processing comprises applying at least one high-confidence model to at least one selected portion of the measured data, wherein the selected portion of the measured data corresponds to a predetermined characteristic pattern of the sensing signal of said at least one sensing type, said high confidence model describing a relation between a change in a cognitive state of the individual and a change in said predetermined characteristic pattern of the sensing signal.

[0018] The predetermined characteristic pattern of the sensing signal corresponds to time variation of the sensing signal of the respective sensing type within at least one characteristic time segment identified during a training stage of said high confidence model. The at least one characteristic time segment is characterized by a condition of optimal first and second merit functions of first and second different models, each of said first and second different models describing a relation between a change in a cognitive state of an individual, during said certain task performance, and a change in the respective measured data sensed on the individual, and being different from the other of said first and second models in at least one of the following conditions: sensing type and ground truth data type of models’ training.

[0019] The first and second measured data, considered in the respective first and second different models, may differ in the sensing types comprising, respectively, motion sensing and autonomous nervous system measures. Additionally, or alternatively, the first and second measured data may differ in the ground truth type comprising, respectively, objective ground truth and error-related ground truth.

[0020] In some embodiments, the sensing data of the at least one sensing type comprises motion data indicative of motion performed by the at least one body part or motion performed by an element operated by the at least one body part of the individual. In some embodiments, the sensing data of the at least one sensing type comprises data indicative of autonomous nervous system (ANS) activity.

[0021] The ground truth data type may describe optimal correlation between objective data corresponding to the change of the cognitive state of an individual and the sensing data of the certain sensing type during a given task being performed by the individual.

[0022] In some embodiments, the at least one high-confidence model is created by carrying out the following: providing a first train set of motion -related measured data MDlSM(t) indicative of motion patterns measured over time on at least one individual using sensing data of each of at least one sensing type, and a second train set of ANS-related measured data MDlANS(t) indicative of simultaneously measured ANS measures on said at least one individual over said time; applying model -based processing to the first train set data MDlSM(t) and the second train set data MD1 ANS(t), the model -based processing comprising: applying to the motion patterns of the first train set data MDlSM(t) a motion-based model M1SM describing a relation between a change in the cognitive state of an individual and a change in measured motion patterns, corresponding to sensing data of each of at least one sensing type, affected by said change in the cognitive state; applying to the measured ANS patterns of the second train set data MDlANS(t) an ANS-based model MIANS describing a relation between a change in the cognitive state of an individual and a change in ANS measures affected by the change in the cognitive state; and identifying in the first train set data MDlSM(t) at least one first characteristic time segment, which is characterized by a characteristic motion pattern having a best match with objective-type ground truth data in relation to the cognitive state of the individual and which at least partially overlaps with at least one second characteristic time segment of a characteristic ANS pattern in the ANS measures of the second train set data MDlANS(t) characterized by a best match with the objective-type ground truth data, thereby identifying at least one pair of the first and second characteristic time segments of matching characteristic motion and ANS patterns, respectively; and creating the at least one high confidence model describing at least one of the following in relation to the objective ground truth data: (i) the relation between a change in the cognitive state of the individual and a change in the characteristic motion pattern; and (ii) the relation between a change in the cognitive state of the individual and a change in the characteristic ANS pattern.

[0023] In some other embodiments, the at least one high confidence model is created by carrying out the following: providing preliminary trains set data comprising a first train set of motion-related measured data MDlSM(t) indicative of motion patterns measured over time on at least one individual, and a second train set of ANS-related measured data MDlANS(t) indicative of simultaneously measured ANS measures on said at least one individual over said time; applying model -based processing to the first train set data MDlSM(t) and the second train set data MD1 ANS(t), the model -based processing comprising: applying to the motion patterns of the first train set data MDlSM(t) a preliminary motion-related model M1SM describing a relation between detection of error in individual’s brain and a change in measured motion patterns affected by said error with respect to error-related ground truth type data, and applying to the ANS measures of the second train set data MDlANS(t) a preliminary ANS- related model MIANS describing a relation between detection of error in the individual’s brain and a change in the ANS measures affected by said error with respect to the corresponding error-related ground truth type data; and identifying in the first train set data MDlSM(t) at least one first characteristic time segment TSM, which is characterized by an error-related characteristic motion pattern having a best match with the error-related ground truth type data in relation to the cognitive state and which is at least partially overlapping with at least one second characteristic time segment TANS of an error-related characteristic ANS pattern of the second train set data MD1 ANS(t) characterized by a best match with the error-related ground truth type data in relation to the cognitive state, thereby identifying at least one pair of first and second characteristic time segments of matching characteristic motion and ANS patterns, respectively, in relation to the error-related ground truth type data; and creating first and second high-confidence models describing, respectively, a first relation between a change in the cognitive state of the individual and a change in the characteristic motion pattern, and a relation between a change in the cognitive state of the individual and a change in the characteristic ANS pattern, in relation to the error-related ground truth type data.

[0024] The above-described model-based processing may further comprise: utilizing error-related first and second train sets comprising, respectively, the error-related characteristic motion patterns and the error-related characteristic ANS patterns, and while parametrically manipulating cognitive state of individuals in relation to objective ground truth data, recording changes in the motion and ANS patterns of said first and second train sets; and creating first and second high-confidence models describing, respectively, a first relation between error commission motion data and the cognitive state of the individual and a second relation between error commission ANS data and the cognitive state of the individual.

[0025] According to another broad aspect of the present disclosure, it provides a method of creating a high-confidence model for use in monitoring a cognitive state of an individual while performing a certain task, the method comprising: measuring evolution of motion patterns over time being indicative of a sensing signal originated on at least one body part of an individual, and simultaneously measuring evolution of autonomous nervous system (ANS) measures of the individual over said time, while intentionally and controllably manipulating the cognitive state of the individual, and recording first and second measured data MDsM(t) and MDANs(t) indicative of, respectively, said evolution of the motion patterns and said evolution of the ANS measures; applying model-based processing to the first and second measured data MDsM(t) and MDANs(t) using, respectively, first and second predetermined models MMand M^NS, said first predetermined model describing a first relation between a change in a cognitive state of the individual and a change in the first measured data MDsM(t) affected by said change in the cognitive state, and the second predetermined model describing a second relation between a change in the cognitive state of the individual and a change in the second measured data MDANs(t) affected by the change in the cognitive state, said model-based processing providing data indicative of at least one pair of first and second characteristic time segments of the first and second measured data, respectively, such that said first and second characteristic time segments are at least partially overlapping, and correspond to first and second characteristic patterns being first and second measured data portions, respectively, characterized by best match conditions of the first and second measured data with ground truth data in relation to the cognitive state defined by the first and second models; creating a high-confidence model describing at least one of the following (i) a relation between a change in the cognitive state of the individual and a change in the motion patterns within the at least one first characteristic time segment; and (ii) a relation between a change in the cognitive state of the individual and a change in the ANS measures within the at least one second characteristic time segment.

[0026] According to yet further broad aspect of the present disclosure, it provides a method of creating a high-confidence model for use in monitoring a cognitive state of an individual while performing a certain task, the method comprising: measuring evolution of motion patterns over time being indicative of a sensing signal originated on at least one body part of an individual and simultaneously measuring evolution of autonomous nervous system (ANS) measures of the individual over said time, while intentionally and controllably inducing error commission by the individual, and recording first motion-related measured data and first ANS-related measured data and MD1ANs(t) indicative of, respectively, said evolution of the motion patterns and said evolution of the ANS measures; applying model-based processing to the first motion-related measured data and said first ANS-related measured data MD^Ns ) using, respectively, first predetermined motion-related model MMand first predetermined ANS-related model M^NS, said first predetermined motion -related model MMdescribing a relation between detection of error commission of the individual and a change in the first motion-related measured data affected by the detection of the error commission, and the first predetermined ANS-related model M^NSdescribing a relation between detection of error commission of the individual and a change in the first ANS-related measured data MD^Ns ) affected by the detection of error commission, said model-based processing providing data indicative of at least one pair of first motion-related and first ANS-related characteristic time segments, TsM=(ti,t2)sM and TANS=(ti,t2)ANS of the first motion-related and first ANS-related measured data, respectively, such that said first motion-related and first ANS-related characteristic time segments are at least partially overlapping, and correspond to first motion-related and first ANS-related characteristic patterns being first motion-related and first ANS-related measured data portions, MD2SM(TSM) and / D^s TANs , respectively, characterized by best match conditions with respective ground truth data in relation to the error commission of the individual defined by the first motion-related and first ANS-related predetermined models; creating two respective models, comprising a motion-related model MMand an ANS-related model MNS, both being configured to detect error commission by the individual, by utilizing only the measured data patterns of said at least one pair of first motion-related and first ANS-related characteristic time segments, TsM=(ti,t2)sM and TANS=(tl,t2)ANS; measuring evolution of motion patterns over time originated on at least one body part of an individual and simultaneously measuring evolution of autonomous nervous system (ANS) measures of the individual over said time while intentionally and controllably manipulating the cognitive state of the individual, and recording second motion-related and second ANS-related measured data MD^SM^) and MD2ANs(t) indicative of, respectively, said evolution of the motion patterns and said evolution of the ANS measures; utilizing second motion-related and second ANS-related characteristic patterns being second motion-related and second ANS-related measured data portions, MD2SM(TSM) and MD2ANS( ANS to create a high-confidence model describing at least one of the following (i) a relation between a change in the cognitive state of the individual and a change in the motion patterns within the at least one motion-related characteristic time segment TSM; and (ii) a relation between a change in the cognitive state of the individual and a change in the ANS measures within the at least one ANS-related characteristic time segment TANS.

[0027] The above methods can be implemented by the above-described control system. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to better understand the subject matter that is disclosed herein and to exemplify how it may be carried out in practice, embodiments will now be described, by way of non-limiting examples only, with reference to the accompanying drawings, in which:

[0029] Fig- 1 is a block diagram showing schematically a system configured according to the principles of the present disclosure for monitoring a cognitive state of an individual while performing a certain task;

[0030] Fig- 2 shows, by way of a flow diagram, the method of the present disclosure using at least one high-confidence model, configured according to the present disclosure and describing a relation between a change in the cognitive state of an individual and a change in measured data, collected from the individual over time, MD(t), within at least one characteristic time segment of the measured data;

[0031] Figs. 3 A to 3C exemplify the model training stage to create the high-confidence model (Fig. 3A), and the use of the high-confidence model to determine the cognitive state of an individual based on measured movement kinematics (Fig. 3B) and ANS measures (Fig. 3C), where the training is performed in relation to respective objective ground truth data / patterns;

[0032] Figs. 4 A to 4C exemplify the model training stage to create the high-confidence model (Fig. 4A) and use of the high-confidence model to determine the cognitive state of an individual based on measured movement kinematics (Fig. 4B) and ANS measures (Fig. 4C), where the training is performed in relation to error-related ground truth data, based on error-related EEG (Electroencephalogram) patterns, behavioral measures, or ANS activity indicating error commission;

[0033] Figs. 5A and 5B exemplify the method of the present disclosure for determining the cognitive state of an individual based on measuring sub-movements only (movement kinematics), wherein Fig. 5A describes models' training to create the high-confidence model, and Fig. 5B describes use of the high-confidence model;

[0034] Figs. 6A and 6B exemplify the method of the present disclosure for determining the cognitive state of an individual based on measuring ANS measures only, wherein Fig. 6A describes models' training to create the high-confidence model, and Fig. 6B describes use of the high-confidence model;

[0035] Fig- 7 shows an example of error-nonrelated sub-movement / ANS patterns / features related to some (manipulated) ground truth cognitive state (BAC / KSS), measured simultaneously showing partially overlapping time segments; and

[0036] Fig- 8 shows an example of error-related sub-movement / ANS patterns / features related to some (manipulated) ground truth cognitive state (BAC / KSS), measured simultaneously showing partially overlapping time segments; the EEG ground truth measured signal is shown as well indicating the appearance of an error-related negativity (ERN).

[0037] DETAILED DESCRIPTION OF EMBODIMENTS

[0038] Reference is made to Fig. 1 showing schematically a system 100 configured and operable according to the present disclosure for monitoring the cognitive state of an individual while performing a certain task. The system 100 is configured for data communication (using any known suitable communication utilities and protocols) with measured data provider(s) and includes a control system 110 configured as a computer system, including data input utility 112, data output utility 114, a memory 116, and a data processor and analyzer 118. The data processor and analyzer 118 includes a measured data analyzer 120 and a cognitive state detector 122.

[0039] The control system 110 is configured to obtain, from at least one measured data provider, e.g., 130-1, measured data, MD(t) over time, indicative of sensing data of at least one sensing type corresponding to a sensing signal originated on at least one body part of the individual and being collected during a sensing time period. The measured data being provided is also typically indicative of respective measuring conditions, i.e., the type of sensing data, e.g., sensor(s) being used, individual’s body part on which the sensing data is collected (the sensing signal is originated), etc.

[0040] The data processor and analyzer 118 is configured and operable for processing the input measured data MD(t) to determine the cognitive state of the individual. The processing includes applying at least one high-confidence model ML to at least one selected portion, MDi(t), of the measured data MD(t). The measured data analyzer 120 is configured and operable to select, from the measured data being provided, the at least one portion MDi(t), which is selected as corresponding to a characteristic pattern of the sensing signal being sensed. As will be described further below, this characteristic pattern is determined / identified, during the model training stage, as corresponding to a predetermined characteristic time segment.

[0041] The at least one high-confidence model ML describes a relation between a change in the cognitive state of the individual and a change in the measured data MD(t). The cognitive state detector 122 operates to extract data indicative of the cognitive state and generate corresponding output data.

[0042] The present disclosure provides the high-confidence model ML optimized to detect the cognitive state of an individual in the sense that the high-confidence model ML is based on first and second different models, describing first and second relations between the cognitive state of the individual and first and second measured data, respectively, collected over said time. The first and second models, while being trained simultaneously, are trained differently, such that the first and second relations, related to the same cognitive state of the same individual, are independent from one another as relating to different and independently collected first and second measured data over the same time. The model training stage is aimed at determining / identifying, for each of the first and second measured data, the at least one first characteristic pattern and at least one second characteristic pattern, respectively, where the first and second characteristic patterns on the one hand correspond to the best fit conditions of the first and second models, respectively, and on the other hand are collected during overlapping first and second time segments, respectively, in the first and second measured data. These time segments thus present the characteristic time segments.

[0043] It should be understood that the measured data is in the form of sensed signal varying over time presenting time variation pattern of the signal. The inventors have found that in some cases each one of the first and second models may utilize one or more respective pattern segments (features) of the respective measured data which are not informative enough with respect to the cognitive condition and therefore, using any of these two models separately may introduce inaccuracies in the determination of the cognitive state of the individual. The goal of the technique of the present disclosure is, on the one hand, eliminate or significantly reduce the less informative information, and on the other hand increase the confidence level of detection of the cognitive state, which is achieved due to extraction of the cognitive state using the high-confidence model being trained based on combination of independent characterization of the cognitive state by different first and second models. The use of the most informative pattern segment(s) of the measured data simplifies and significantly improves the accuracy of cognitive state detection during real time.

[0044] Thus, the at least one characteristic pattern segment of the measured data is characterized by best fit conditions of the first and second measured data with respect to the first and second different models, respectively, i.e., conditions of optimal first and second merit functions of the first and second different models. These first and second different models describe a change in the cognitive state of an individual during a certain task affecting a change in the first and second measured data, respectively, being sensed on the individual, while being different in that the first and second models correspond to first and second conditions differing in sensing type of the measured data and / or ground truth data type used in models’ training.

[0045] More specifically, the first and second models are trained to describe relations between the specific cognitive state (ground truth based) and, respectively, measured motion pattern over time (first measured data) and ANS measures over time (second measured data). The ground truth may be of the objective type (e.g., BAC / KSS), or may be of the error-related type. In the following description the objective ground truth type (e.g., BAC / KSS) is denoted GT1, and the error-related ground truth type is denoted GT2.

[0046] It should be noted that, during training stages of models used in the present disclosure, first and second different measured data being used and corresponding to different sensing types, e.g., motion patterns and ANS measures, may utilize the same ground truth type (GT1 or GT2), as will be described in detail further below.

[0047] Reference is made to Fig. 2 showing, by way of a flow diagram 200, the method of the present disclosure of using the at least one high-confidence model ML described above. In step 210, measured data MD(t) is obtained in real time, and is indicative of sensing data of at least one sensing type, corresponding to a sensing signal originated on at least one body part of an individual performing a certain task (under certain activity) and being collected over time t from said body part or an element / device operated by / associated with said body part (presenting a measurement condition). As noted above, the measured data includes or is accompanied with data indicative of the sensing type and of the measurement condition. For example, the sensing data may include the time profile of the motion pattern, e.g., measured steering wheel angle of an individual while driving a car over time t. It should be noted that alternatively or additionally, the sensing data may be ANS data.

[0048] In step 220, the at least one high-confidence model ML is applied to at least one selected portion MDi(t) of the measured data (i.e., indicative of the sensing data corresponding to the sensing signal collected during a characteristics time segment t, of the sensing time f) having the predetermined characteristic pattern of the high-confidence correspondence of a change in the cognitive state of the individual (e.g., drunkenness or fatigue) and the features of the motion pattern. For example, the high-confidence model ML may describe the relationship between a change in the drunkenness level of the individual and a change in a pattern of submovements detected in the measured motion pattern of the steering wheel angle profile. In step 230, the cognitive state of the individual (e.g., the level of drunkenness) is reported in real time.

[0049] In the following, specific examples of training and use of the at least one high- confidence model ML are exemplified in detail.

[0050] Reference is made to Figs. 3A to 3C showing, respectively, an example of a model training procedure to thereby create high-confidence model(s) (Fig. 3A) and the use of the trained high-confidence models to determine the cognitive state (Figs. 3B and 3C). In this example, the first and second measured data are indicative of, respectively, movement kinematics and ANS measures, and the model training is performed with respective to objective ground truth patterns GT1.

[0051] As described above, the task performed by individual(s) during the model training procedure is identical to the real time task during which the user’s cognitive state is to be monitored / determined.

[0052] Fig. 3A exemplifies the model training stage. During training, individual’s cognitive state (e.g., drunkenness or fatigue) is manipulated and both the movement kinematics and the ANS measures are measured over time t (within a certain time interval). At least two predetermined models, MMand M^NSare used to detect cognitive states, wherein the first model is based on sub-movements (generally, movement kinematics) and the second model M^NSis based on ANS measures. In other words, the first model M M describes a first relation between changes in the measured motion data / pattems and changes in the cognitive state of the individual (affecting said changes in the measured motion data), for the given type of sensors (i.e., sensing data types) being used to collect the sensing signal to obtain the respective measured motion patterns. Similarly, the second model M^NSdescribes a second relation between changes in the ANS measures and changes of the cognitive state of the individual (affecting said changes in the ANS measures), for the given type of sensors (i.e., sensing data types) being used to collect the sensing signal to obtain the ANS measures.

[0053] It should also be noted that the training stage, with respect to a given task performance, may include various pairs of first and second models previously provided in relation to collection of sensing data using various types of sensors, respectively.

[0054] Motion patterns (e.g., steering wheel angle during driving a car) are measured by motion sensor(s) over time t and analyzed over the respective ground truth data GT1 (objective-type ground truth data) to train the first model MM. For example, the ground truth data GT1 (sub-movement patterns) may be previously determined by manipulation of blood alcohol concentration (BAC) in case of drunkenness or of fatigue level according to Karolinska Sleepiness Scale (KSS).

[0055] One or more pattern segments / portions from the first measured data (motion data) (shown here as a sequence of pattern segments collected during time windows A, B, C... E) are identified (termed here "first pattern segments" as relating to the first model), as being the pattern segments having the best match between the measured and ground truth GT1 data, e.g., first pattern segments MDSM(A) and MDSM(E).

[0056] Similarly, ANS measures are simultaneously collected by ANS sensor(s) over time t (the same time, i.e., simultaneous training of both models) and analyzed over the respective ground truth pattern GT1 to train the second model M^NS. One or more pattern segments / portions from the second measured data (the ANS measures) (shown here as a sequence of measures collected during time segments A', B', C) are identified, presenting second pattern segments having the best match with the ground truth GT1, e.g., pattern segments MDANS(A') and MDANS(B').

[0057] It should be noted that the length of any of the time segments A, B, C. . . E may or may not be equal to the length of any of the time segments A', B', C.

[0058] The first pattern segments MDSM(A) and MDSM(E) and the second pattern segments MDANS(A') and MDANS(B') of the best match with the theoretical data (ground truth data GT1) are analyzed to select at least one pair of at least partially overlapping first and second corresponding time segments, e.g., A and A' time segments in this example. The respective first and second pattern segments MDSM(A) and MDANS(A') are characteristic pattern segments selected as optimally characterizing the cognitive condition while measured by respective sensors in relation to a given task performance.

[0059] Based on the above training stage, two respective high-confidence models are created, one (first) model, MM, is based on sub-movements (generally, kinematic data) and the other (second) model, MNS, is based on ANS measures. These two models describe respective relations between respective measured data and optimized theoretical data, being optimized data indicative of characteristic measured patterns, MDSM(A) and MDANS(A'), identified as the best-fit patterns collected during at least partially overlapping time segments A and A' (characteristic time segments) during simultaneous motion-based and ANS-based measurements. Both models, MMand M^NSprovide detection of the cognitive state with best match as indicated by the respective patterns (i.e., sub-movements and ANS measures) of the ground truth data GTE

[0060] It should be noted that, generally, according to the technique of the present disclosure (relevant for any of the examples in the present disclosure described herein), during the model training stage implemented in relation to a specific task performance, more than one initial model MMcan be used, where each such motion-related initial model is associated with a specific type of sensing data used for motion patterns detection under specific measurement condition. Similarly, more than one initial model M^NScan be used, where each such ANS-related initial model is associated with a specific type of sensing data used for ANS measures detection. The initial models are previously prepared and stored in a memory, e.g., memory 116 of system 100 or that of an external station (server), and properly accessed and used to perform the training stage to create high-confidence models.

[0061] Thus, each high-confidence model relates to specific performance task, type of measured data (kinematic data or ANS data), type of sensing data used under specific measurement conditions to provide the measured data. Hence, various training stages can be performed to create a plurality of high-confidence models which are properly stored in a database (e.g., managed by a server system), enabling the system 100 to select the relevant high-confidence model(s) for processing the relevant measure data collected using relevant sensor types, to determine the cognitive state / condition of an individual during relevant task performance.

[0062] Figs. 3B and 3C describe how each of the high-confidence models and 2

[0063] MA Scanbe used for actual (real-time) detection of the cognitive state of an individual during the given task performance.

[0064] In some embodiments (Fig. 3B), during the real time state, only motion sensor(s) is / are used (of the same type as used for the model training). Measured data MDsM(t) is acquired (or provided / received from measured data provider) being indicative of motion patterns collected / sensed over time using specific sensor type(s). The high-confidence model, M M, extracts the measured data portion(s) corresponding to the characteristic pattern (being the motion signal pattern corresponding to the best-match signal pattern collected during the above-described characteristic time segment A identified during the training and serving the basis of the high-confidence model MMcreation), and generates data about the cognitive condition.

[0065] In some embodiments (Fig. 3C), only ANS measures are used (of the same type as used for training). Measurement data MD NS(I) indicative of ANS measures is acquired over time (or provided from measured data provider). The high-confidence model, MA S? extracts the measured data portion corresponding to the characteristic signal pattern (being the ANS signal pattern corresponding to the above-described bestmatch ANS pattern segment acquired during the characteristic time segment A' identified 2 during the training and serving the basis of the high-confidence model M^NScreation), and generates data about the cognitive condition.

[0066] In some other embodiments, both motion sensor and ANS measures are used (of the same types as used for training). Respective measured data MDsM(t) and MDANs(t) are simultaneously collected over time or respective data are obtained from the measured data provider. The respective high-confidence models MMand MNSidentify the characteristic pattern segments (having respective characteristic features) corresponding to those of the characteristic time segments A and A' identified during the training procedure, and generate data about the cognitive condi tion / state of the individual.

[0067] Reference is made to Figs. 4A to 4C showing, respectively, an example of model training procedure to create at least one high-confidence model and use of the at least one high-confidence model to determine the cognitive state of an individual based on movement kinematics, ANS measures and error-related ground truth datatype. Such error-related ground truth data may be determined based on EEG (Electroencephalogram) patterns, behavioral measures (e.g., motion patterns), or ANS activity indicating error commission. As noted above, this type of ground truth data is denoted in the following as GT2.

[0068] In electrophysiological research, the error-related negativity (ERN) is defined as a negative voltage deflection peaking at fronto-central electrode sites (Fz, FCz, Cz) approximately 50-100 msec after the commission of an error and is known to correlate with error processing in the brain. This potential is commonly attributed to reflect activity of the areas implicated in error monitoring. The second component of the error-related brain potential is known as error positivity (Pe) which follows the ERN in time, and peaks approximately 200-400 ms after a performance error occurs. Additional error-related brain potentials that can be used as ground truth data GT2 for 'error' instead of or in addition to ERN include Feedback Related Negativity (FRN), Medial Frontal Negativity (MFN), and Conflict-related negativity (N2). In addition to the mentioned above error- related brain potential components belonging to the time domain of the EEG, frequency domain components of the EEG waves can also be used as a ground truth in the techniques of the present disclosure. Examples of such EEG waves related to the frequency domain include alpha POWER and theta POWER because they were shown to appear when a person is under cognitive load or drunk.

[0069] Previous work has demonstrated that errors also give rise to a strong response of the autonomic nervous system (ANS). For example, relative heart rate deceleration was reported to occur after errors, and changes in skin conductance and pupil diameter were shown to be associated with error commission.

[0070] Fig. 4A exemplifies the training stage when detection of ERN serves as the ground truth. During the model training procedure, motion-related measured data MDsM(t) is acquired including motion patterns being indicative of sensing data (for each of at least one sensing type) corresponding to a sensing signal collected / sensed over time t ,and this motion-related measured data MDsM(t) is analyzed over respective ground truth patterns GT2 (error-related ground truth data) to train the first initial model MM. Thus, model MMis trained to identify sub-movements and error-related data (e.g., ANS patterns) arising following error detection in the user’s brain. As already noted above, here, error- related EEG patterns, behavioral measures, or ANS activity indicating error commission, may serve as an error-related ground truth type data (denoted GT2). It should be noted that, alternatively, such first initial model MMcan be previously provided and stored, and during the training stage this first initial model MMis properly accessed.

[0071] Detection of characteristic motion pattern (submovements) corresponding to error-related phenomena may be examined in different time segments (e.g., A, B, C, D) depending on the specific characteristics of the error-related phenomena of interest. One or more pattern segments from segments A, B, C, D are identified, being pattern segments having the best match with the ground truth GT2 (i.e., correct error detection of individual), e.g., pattern segments MDSM(B) and MDSM(C).

[0072] For example, the error-related negativity (ERN), i.e., the EEG potential associated with error initiation and / or correction may be searched for by using a variety of “singletrial” analysis methods known to experts in this field. Error-related brain potential components of the EEG are commonly investigated in the literature on brain activity that is averaged across multiple data samples (e.g., averaging the signals of several error commission occurrences). Nevertheless, there are also techniques to find error-related EEG components in a single occurrence, without averaging. Alternatively, if the timing of the execution of the error itself is available, for example from an analysis of the results of the user's actions (e.g., crossing and returning to lane, pressing the breaks) or specific movement kinematics, the ERN may be searched for within 200 ms before, and / or 300ms after the error indication, while the Pe (EEG potential associated with error awareness) may be searched for within 100 to 600 ms after the error indication. In addition, peak theta amplitude (an error-related component in the frequency domain) may be measured between -400 and 600 ms with respect to error indication [ Kieffaber PD. et al. “Deconstructing the functional significance of the error- related negativity (ERN) and midline frontal theta oscillations using stepwise timelocking and single-trial response dynamics”. NeuroImage 274: 120113, (2023)]. These brain potential components (ERN and Pe) are examined with central electrodes (e.g., FCz, Fz, Cz).

[0073] The ANS patterns are measured by ANS sensor(s) over the same time t (i.e., simultaneous training of both models) and analyzed over respective ground truth patterns GT2 (corresponding to error detection of individual) to train a second initial model M^NS. Similarly, such first initial model M^NScan be previously provided and stored, and during the training stage this first initial model M^NSis properly accessed.

[0074] One or more characteristic pattern segments from time segments A', B' are identified, being the pattern segments having the best match with the ground truth GT2, e.g., characteristic pattern segments MDANS(A') and MDANS(B'). The time segments known to be sensitive to error commission may be in 500 ms time windows [Wessel JR et al., “Error Awareness Revisited: Accumulation of Multimodal Evidence from Central and Autonomic Nervous Systems”, Journal of Cognitive Neuroscience 23:3021-3036, (2011)] within the 3s post-error indication events [ Hajcak G. et al., “To err is autonomic: Error-related brain potentials, ANS activity, and post-error compensatory behavior”. Psychophysiology 40:895-903, (2003)]. Thus, model M^NSis trained to detect error commission by the individual.

[0075] Initial characteristic pattern segments, being motion-based error-related pattern segments MDSM(B) and MDSM(C) and ANS-based error-related pattern segments MDANS(A') and MDANS(B'), are analyzed to select at least one pair from these motionbased and ANS-based error-related patterns which have been collected during at least partially overlapping characteristic time segments (e.g., time segments B and A'), and the respective characteristic pattern segments MDSM(B) and MDANS(A') are selected as optimally characterizing the error commission condition while measured by respective sensors.

[0076] The characteristic patterns MDSM(B) and MDANS(A') corresponding to this at least one pair of at least partially overlapping characteristic time segments (B and A') are used to create two respective high-confidence models, one model, is based on motion data (sub-movements) and the other model, M^NS, is based on ANS measures. These models, MMand MNS, are trained, respectively, on the characteristic time segments B and A' only, which, as described above, correspond to the pattern segments identified by the respective models, MMand M^NSas the best-fit error commission relating patterns with respect to the ground truth GT2.

[0077] The training stage may further proceed towards the creation of two new high-

[0078] 3 3 confidence models, MMand M^NS. This training step is aimed at identifying the changes in the error-related characteristic patterns (sub -movements and ANS, respectively) of the time segments (B and A') chosen above, caused by changes in the cognitive state (e.g., drunkenness or fatigue) of the individual.

[0079] In other words, each of high-confidence models, MMand M^NSdescribes a relation between the error commission measured data and the cognitive of the individual, enabling to relate the error commission to the factor / source of change in the cognitive state leading to the error in the task performance.

[0080] More specifically, the cognitive state (e.g., drunkenness, fatigue) of the individual(s) is parametrically manipulated (e.g., obtained by manipulation of BAC or KSS), respective patterns (i.e., sub-movements and ANS measures) are measured, and the high-confidence models, MMand M^NSare created describing the relations between changes in these characteristic patterns (i.e., sub-movements and ANS measures) and changes in the cognitive state of the individual (based on the objective ground truth data GT1. The high-confidence models MMand M^NSare considered as being error-related sub-movements based (MfM) and error-related ANS activity-based because the 3 3 time segments, B and A', chosen for training of these models, MMand M^NS, satisfied the best-fit condition for error detection of individual. Both high-confidence models, and M^NS, rely on the characteristic patterns / features identified in the previous step of the training stage (creation of MMand M^NS) and built according to the changes in these patterns / features in response to different levels of the cognitive state of interest. It should be understood that first and second pluralities of respective motion-based and ANS-based models MMand M^NScan be provided in relation to various factors / sources affecting the change in the cognitive state.

[0081] It may be stated that, although the objective ground truth GT1 (being error- nonrelated, e.g., BAC or KSS) was used for training and creation of the high-confidence models MSMand M^NS, the ground truth type in this case can be denoted as a conditional one GT1 / GT2. This is because the training sets used for training / creation of the high- confidence models MMand M^NSwere limited to the error-related motion and ANS pattern segments. In other words, the objective ground truth data GT1 was applied only to the characteristic patterns corresponding (identified for the characteristic time segments) which showed best fit with respect to error-related ground truth data GT2 2 2 during creation of models MMand M^NS.

[0082] Further below, method based on an unconditional conjunction / combination of the two types of ground truth data will be demonstrated.

[0083] Figs. 4B and 4C describe actual detection of the cognitive state of the individual using the high-confidence models MMand M^NS, respectively. After creation (using the model training procedure described above), each of these high-confidence models is applied to the real-time measured data, obtained during relevant task performance from sensors relevant to it.

[0084] For example, in the case of measuring a cognitive state while driving, the motion- related (sub-movement) model (M^M) is applied to motion measured data sensed on the steering wheel, and the autonomous model (MANS) is applied to autonomous activity relating real-time measured data (ANS data) obtained from sensors of autonomous activity such as a camera, radar, or wearable sensors. Reference is made to Figs. 5A and 5B exemplifying the technique, according to the present disclosure, for determining the cognitive state of an individual based on measuring motion patterns (sub-movements) only. The method is based on the inventors' understanding that the certainty / accuracy of cognitive state determination may benefit from an unconditional conjunction (combination) of error-related and error-nonrelated models. In other words, identification of error commission may be uniquely affected by an impaired cognitive state of the individual in real time.

[0085] Fig. 5A exemplifies the training stage when both the error-related ground truth data GT2 (e.g., detection of ERN) and objective ground truth data GT1 of the cognitive state (e.g., BAC or KSS) serve as ground truths. In the first stage, an initial model, MM, is trained to identify characteristic motion-based patterns accompanying error detection in the user’s brain.

[0086] Specifically, as described above with reference to Fig. 4A, analysis of the individual’s movement kinematics (sub-movements) at time segments / windows close to error detection-related phenomena (e.g., ERN) is performed over respective ground truth patterns GT2 (error-related ground truth data). Here, error-related EEG patterns or behavioral measures indicating error commission or other ANS activity, serve as the ground truth GT2. To this end, error detection-related phenomena may be examined in different time segments (e.g., A, B, C, D) depending on the specific characteristics of the error-related phenomena of interest.

[0087] In the next stage, a cognitive state is parametrically manipulated (e.g., by using various levels of BAC or KSS) and first and second models, MMand MMare trained. The first model, MM, is based on the analysis of the characteristic sub-movement patterns that were identified during creation of the initial model MM, as being related to error commission. This model MMis actually taught to detect changes in errordependent features / patterns as a result of changes in the cognitive state (e.g., drunkenness or fatigue) utilizing the patterns of the same time segments as model MMand identifying the characteristic time segments (e.g., A, C, D) in which the patterns of the measured data have the best match with the objective ground truth GT1, e.g., BAC or KSS (being error- related and denoted as GT1 / GT2) . The second model MMis taught to identify changes in the movement patterns (regardless of making an error) associated with the change in the cognitive state based on the objective ground truth GT1 (objective ground truth data accepted in the field), being substantially error-nonrelated (e.g., BAC / KSS).

[0088] Thus, the second model MMdescribes the relation between the characteristic patterns of different characteristic time segments (e.g., E, C, D) having the best match with the objective ground truth GT1.

[0089] In the final stage, a new high-confidence model, MM, is created that describes the relation between changes in the cognitive state and variation of the characteristic movement patterns corresponding to those of the characteristic time segments / windows 2

[0090] (e.g., C and D) in which the two models (the error detection-related model) and MM(the “error agnostic model”) agreed about the cognitive state, i.e., had a best match with the objective ground truth data GT1.

[0091] Fig. 5B describes that, once the new high-confidence model, MM, is formalized, it is used in real-time, i.e., is applied to real-time measured data obtained from sensors relevant to it, i.e., motion data (e.g., sub-movements). For example, in the case of measuring a cognitive state while driving, the sub-movement-based high-confidence model MMis applied to measured data from the steering wheel and detects a possible impaired cognitive state of the individual.

[0092] Reference is made to Figs. 6A and 6B exemplifying the technique, according to the present disclosure, for determining the cognitive state of an individual based on measuring ANS measures only. Here again, the technique is based on that certainty / accuracy of cognitive state determination may benefit from an unconditional conjunction (combination) of error-related and error-nonrelated models. In other words, identification of error commission may be associated with an impaired cognitive state of the individual in real time.

[0093] Fig. 6A exemplifies the training stage when both the ground truth data GT2 of error commission (e.g., detection of ERN) and objective ground truth data GT1 of the cognitive state (e.g., BAC or KSS) serve as ground truths. In the first stage, an initial model, M^NS, is trained to identify characteristic ANS patterns accompanying error detection in the user’s brain. Specifically, as described above with reference to Fig. 4A, analysis of individual’s ANS measures at time segments / windows close to error detection-related phenomena (e.g., ERN) is performed over respective ground truth patterns GT2 (error-related ground truth data). Here, error-related EEG patterns or behavioral measures indicating error commission or other ANS activity, serve as the ground truth GT2. To this end, error detection-related phenomena may be examined in different time segments (e.g., A, B, C, D) depending on the specific characteristics of the error-related phenomena of interest.

[0094] In the next stage, a cognitive state is parametrically manipulated (e.g., by using various levels of BAC or KSS) and two models, MANS and M^NS are trained. First model, MNS, is based on the analysis of the characteristic ANS patterns that were identified during creation of model MANS?asbeing related to an error commission. This model MA S) is actually taught to detect changes in error-dependent features / pattems as a result of changes in the cognitive state (e.g., drunkenness or fatigue) utilizing the patterns of the same time segments as model MANSar|d identifying the characteristic time segments (e.g., A, C, D) in which the patterns of the measured data have the best match with the ground truth GT1, e.g., BAC or KSS (being error-related and denoted as GT1 / GT2) . The second model, M^NS? is taught to identify changes in the ANS patterns (regardless of making an error) associated with the change in the cognitive state based on ground truth GT1 (objective ground truth data accepted in the field), being substantially error-nonrelated (e.g., BAC / KSS).

[0095] Thus, the second model, MANS> describes the relation between the characteristic patterns of different characteristic time segments (e.g., E, C, D) having the best match with the ground truth GTE

[0096] In the final stage, a new high-confidence model, MANS? is trained / created that describes the relation between changes in the cognitive state and variation of the characteristic ANS patterns corresponding to those of the characteristic time segments / windows (e.g., C and D) in which the two models MANS (the error detection- related model) and M^NS (the “error agnostic model”) agreed about the cognitive state, i.e., had a best match with the objective ground truth data GT1. Fig. 6B indicates that once the high-confidence model, MNS, isformalized, it is used in real-time, i.e., is applied to real-time measured data obtained from sensors relevant to it, i.e., measuring heart rate. For example, in the case of measuring a cognitive state while driving, the ANS-based model M^NSis applied to measured data from sensors of autonomous activity such as camera, radar, or wearable sensors, and detects a possible impaired cognitive state of the individual.

[0097] Fig. 7 shows an example of three types of measured data MDl(t), MD2(t), MD3(t) obtained simultaneously over a period of 3 secs during driving activity of the individual. In this example the measured data MDl(t) is indicative of steering wheel angles, and measured data MD2(t) is indicative of heart rate signals. The third curve, MD3(t) corresponds to EEG signal and shows that no error was detected, since no ERN signal is detected. Such error-nonrelated sub-movement / ANS pattems / features measured simultaneously may be related to some (manipulated) ground truth cognitive state GT1 (BAC / KSS). Some of the features in the steering wheel angles and heart rate signals may appear in partially overlapping time segments.

[0098] Fig. 8 shows another example of three types of measured data MDl(t), MD2(t), MD3(t) obtained simultaneously over a period of 3 secs. As in the example of Fig. 7, the measured data MDl(t) is indicative of steering wheel angles, measured data MD2(t) is indicative of heart rate signals. The third curve, MD3(t), corresponds to EEG signal and shows that an error was detected, indicated by a strong ERN signal detected simultaneously with the steering wheel angles and heart rate signals.

[0099] Typically, the selected time window for detecting characteristic motion patterns may be 1 to 1.5 seconds preceding the local minimum of the ERN. The selected time window for characteristic ANS pattems / measures may be 0 to 3 seconds (depending on the ANS measure) following the local minimum of the ERN.

[0100] It should be noted that in the error-related case, some of the features in the steering wheel angles and heart rate signals appear in partially overlapping time segments (Fig. 8). When there is no ERN, the time relation between ANS and steering is more variable.

Claims

CLAIMS:

1. A system for monitoring a cognitive state of an individual during a certain task performance, the system comprising: a control system configured as a computer system comprising data input and output utilities, a memory, and a data processor and analyzer, the control system being configured and operable for processing input measured data, indicative of time variation of sensing data of at least one sensing type corresponding to a sensing signal being originated on at least one body part of the individual over time under predetermined measuring conditions, to determine the cognitive state of the individual, wherein said processing comprises applying at least one high-confidence model to at least one selected portion of the measured data, wherein the selected portion of the measured data corresponds to a predetermined characteristic pattern of the sensing signal of said at least one sensing type, said high confidence model describing a relation between a change in a cognitive state of the individual and a change in said predetermined characteristic pattern of the sensing signal.

2. The system according to claim 1, wherein said predetermined characteristic pattern corresponds to time variation of the sensing signal of the at least one sensing type within at least one characteristic time segment identified during a training stage of said high confidence model, said at least one characteristic time segment being characterized by a condition of optimal first and second merit functions of first and second different models, each of said first and second different models describing a relation between a change in a cognitive state of an individual, during said certain task performance, and a change in the measured data sensed on the individual, and being different from the other of said first and second models in at least one of the following: sensing type and ground truth data type of models training.

3. The system according to claim 1 or 2, wherein said sensing data of the at least one sensing type comprises motion data indicative of motion performed by the at least one body part or motion performed by an element operated by the at least one body part of the individual.

4. The system according to claim 1 or 2, wherein said sensing data of the at least one sensing type comprises data indicative of autonomous nervous system (ANS) activity.

5. The system according to claim 1 or 2, wherein said sensing data of the at least one sensing type comprises motion data indicative of motion performed by the at least one body part of the individual or motion performed by an element operated by the at least one body part of the individual, and data indicative of autonomous nervous system (ANS) activity of the individual.

6. The system according to any one of claims 2 to 5, wherein said first and second different models utilize different first and second sensing types comprising, respectively, motion sensing and autonomous nervous system measures.

7. The system according to any one of claims 2 to 5, wherein said first and second different models utilize ground truth data of different ground truth types comprising error- related ground truth and objective ground truth.

8. The system according to claim 6, wherein said first and second different models utilize ground truth data of different ground truth types comprising error-related ground truth and objective ground truth.

9. The system according to any one of claims 2 to 8, wherein said at least one high confidence model is created by carrying out the following: providing a first train set of motion -related measured data MDlSM(t) indicative of motion patterns measured over time on at least one individual using sensing data of each of at least one sensing type, and a second train set of ANS-related measured data MDlANS(t) indicative of simultaneously measured ANS measures on said at least one individual over said time; applying model -based processing to the first train set data MDlSM(t) and the second train set data MD1 ANS(t), the model -based processing comprising: applying to the motion patterns of the first train set data MDlSM(t) a motion-based model M1SM describing a relation between a change in the cognitive state of an individual and a change in measured motion patterns, corresponding to sensing data of each of at least one sensing type, affected by said change in the cognitive state; applying to the measured ANS patterns of the second train set data MDlANS(t) an ANS-based model MIANS describing a relation between achange in the cognitive state of an individual and a change in ANS measures affected by the change in the cognitive state; and identifying in the first train set data MDlSM(t) at least one first characteristic time segment, which is characterized by a characteristic motion pattern having a best match with objective-type ground truth data in relation to the cognitive state of the individual and which at least partially overlaps with at least one second characteristic time segment of a characteristic ANS pattern in the ANS measures of the second train set data MDlANS(t) characterized by a best match with the objective-type ground truth data, thereby identifying at least one pair of the first and second characteristic time segments of matching characteristic motion and ANS patterns, respectively; and creating the at least one high confidence model describing at least one of the following in relation to the objective ground truth data: (i) the relation between a change in the cognitive state of the individual and a change in the characteristic motion pattern; and (ii) the relation between a change in the cognitive state of the individual and a change in the characteristic ANS pattern.

10. The system according to any one of claims 2 to 8, wherein said at least one high confidence model is created by carrying out the following: providing preliminary trains set data comprising a first train set of motion-related measured data MDlSM(t) indicative of motion patterns measured over time on at least one individual, and a second train set of ANS-related measured data MDlANS(t) indicative of simultaneously measured ANS measures on said at least one individual over said time; applying model -based processing to the first train set data MDlSM(t) and the second train set data MD1 ANS(t), the model -based processing comprising: applying to the motion patterns of the first train set data MDlSM(t) a preliminary motion-related model M1SM describing a relation between detection of error in individual’s brain and a change in measured motion patterns affected by said error with respect to error-related ground truth type data, and applying to the ANS measures of the second train set data MDlANS(t) a preliminary ANS- related model MIANS describing a relation between detection of error in theindivi dual’s brain and a change in the ANS measures affected by said error with respect to the corresponding error-related ground truth type data; and identifying in the first train set data MDlSM(t) at least one first characteristic time segment TSM, which is characterized by an error-related characteristic motion pattern having a best match with the error-related ground truth type data in relation to the cognitive state and which is at least partially overlapping with at least one second characteristic time segment TANS of an error-related characteristic ANS pattern of the second train set data MD1 ANS(t) characterized by a best match with the error-related ground truth type data in relation to the cognitive state, thereby identifying at least one pair of first and second characteristic time segments of matching characteristic motion and ANS patterns, respectively, in relation to the error-related ground truth type data; and creating first and second high-confidence models describing, respectively, a first relation between a change in the cognitive state of the individual and a change in the characteristic motion pattern, and a relation between a change in the cognitive state of the individual and a change in the characteristic ANS pattern, in relation to the error-related ground truth type data.

11. The system according to claim 10, wherein said model -based processing further comprises: utilizing error-related first and second train sets comprising, respectively, the error-related characteristic motion patterns and the error-related characteristic ANS patterns, and while parametrically manipulating cognitive state of individuals in relation to objective ground truth data, recording changes in the motion and ANS patterns of said first and second train sets; and creating first and second high-confidence models describing, respectively, a first relation between error commission motion data and the cognitive state of the individual and a second relation between error commission ANS data and the cognitive state of the individual.

12. A method of creating a high-confidence model for use in monitoring a cognitive state of an individual while performing a certain task, the method comprising: measuring evolution of motion patterns over time being indicative of a sensing signal originated on at least one body part of an individual, and simultaneously measuringevolution of autonomous nervous system (ANS) measures of the individual over said time, while intentionally and controllably manipulating the cognitive state of the individual, and recording first and second measured data MDsM(t) and MDANs(t) indicative of, respectively, said evolution of the motion patterns and said evolution of the ANS measures; applying model-based processing to the first and second measured data MDsM(t) and MDANs(t) using, respectively, first and second predetermined models MMand M^NS,said first predetermined model describing a first relation between a change in a cognitive state of the individual and a change in the first measured data MDsM(t) affected by said change in the cognitive state, and the second predetermined model describing a second relation between a change in the cognitive state of the individual and a change in the second measured data MDANs(t) affected by the change in the cognitive state, said model-based processing providing data indicative of at least one pair of first and second characteristic time segments of the first and second measured data, respectively, such that said first and second characteristic time segments are at least partially overlapping, and correspond to first and second characteristic patterns being first and second measured data portions, respectively, characterized by best match conditions of the first and second measured data with ground truth data in relation to the cognitive state defined by the first and second models; creating a high-confidence model describing at least one of the following (i) a relation between a change in the cognitive state of the individual and a change in the motion patterns within the at least one first characteristic time segment; and (ii) a relation between a change in the cognitive state of the individual and a change in the ANS measures within the at least one second characteristic time segment.

13. A method of creating a high-confidence model for use in monitoring a cognitive state of an individual while performing a certain task, the method comprising: measuring evolution of motion patterns over time being indicative of a sensing signal originated on at least one body part of an individual and simultaneously measuring evolution of autonomous nervous system (ANS) measures of the individual over said time, while intentionally and controllably inducing error commission by the individual, and recording first motion-related measured data and first ANS-related measured dataMD^M^) and MD1ANs(t) indicative of, respectively, said evolution of the motion patterns and said evolution of the ANS measures; applying model-based processing to the first motion-related measured data MD2sM(t) and said first ANS-related measured data MD^Ns ) using, respectively, first predetermined motion-related model MMand first predetermined ANS-related model M^NS,said first predetermined motion -related model MMdescribing a relation between detection of error commission of the individual and a change in the first motion-related measured data MD2sM(t) affected by the detection of the error commission, and the first predetermined ANS-related model M^NSdescribing a relation between detection of error commission of the individual and a change in the first ANS-related measured data MD^Ns ) affected by the detection of error commission, said model-based processing providing data indicative of at least one pair of first motion-related and first ANS-related characteristic time segments, TsM=(ti,t2)sM and TANS=(ti,t2)ANS of the first motion-related and first ANS-related measured data, respectively, such that said first motion-related and first ANS-related characteristic time segments are at least partially overlapping, and correspond to first motion-related and first ANS-related characteristic patterns being first motion-related and first ANS-related measured data portions, MD^M TSM) and MD^NSCTANS), respectively, characterized by best match conditions with respective ground truth data in relation to the error commission of the individual defined by the first motion-related and first ANS-related predetermined models; creating two respective models, comprising a motion-related model MMand an ANS-related model M^NS, both being configured to detect error commission by the individual, by utilizing only the measured data patterns of said at least one pair of first motion-related and first ANS-related characteristic time segments, TsM=(ti,t2)sM and TANS=(tl,t2)ANS; measuring evolution of motion patterns over time originated on at least one body part of an individual and simultaneously measuring evolution of autonomous nervous system (ANS) measures of the individual over said time while intentionally and controllably manipulating the cognitive state of the individual, and recording second motion-related and second ANS-related measured data MD^SM^) and MD2ANs(t)indicative of, respectively, said evolution of the motion patterns and said evolution of the ANS measures; utilizing second motion-related and second ANS-related characteristic patterns being second motion-related and second ANS-related measured data portions, MD2SM(TSM) and MD2ANS( ANS to create a high-confidence model describing at least one of the following (i) a relation between a change in the cognitive state of the individual and a change in the motion patterns within the at least one motion-related characteristic time segment TSM; and (ii) a relation between a change in the cognitive state of the individual and a change in the ANS measures within the at least one ANS-related characteristic time segment TANS.

Images