Method and system for generating and utilizing data to detect and continuously monitor human body state

Abstract

The invention relates to a method and system for establishing true values in human body state detection. In one embodiment, the disclosure teaches to record physiological data over a continuous duration of at least a threshold duration sufficient to capture a representation of a target state in a biosignal for detection. The recorded data is segmented into shorter analysis windows as a step of proceeding towards persistent state detection, the duration of each window being less than a threshold and marked with a true value indicative of the target state. The windows and tags are stored in non-transitory memory for later use in training, testing, and verifying one or more state prediction models. The method improves the accuracy of the state detection model by providing more typical true values through long-term recordings that capture different manifestations of the body state in the physiological process.

Description

Technical Field

[0001] This disclosure generally relates to the field of human condition detection and monitoring. More specifically, this disclosure relates to generating and utilizing real-value data to train, test, and validate systems for the continuous monitoring of human condition. Background Technology

[0002] In recent years, the development of solutions for detecting and continuously monitoring human states, such as drowsiness, stress, and cognitive load, has attracted considerable interest, particularly in activities where performance and attention are relevant factors, such as driving. Current methods for detecting human states rely on subjective or objective measurements of underlying psychophysiological states, typically utilizing physiological signals recorded from the heart, brain, skin, eyes, and other organs. These signals reflect underlying psychophysiological states that can influence and affect an individual's behavior and performance.

[0003] Despite advancements in mathematical modeling and machine learning / artificial intelligence (ML / AI) capabilities, accurately detecting human states remains a challenging task. The inventors have identified one of the main difficulties in accurately predicting human states as the dynamic and variable nature of the physiological manifestations of underlying psychophysiological states. For example, the manifestations of human states may occur intermittently and may be interspersed with various behaviors and mental activities associated with the subject, but are not always directly related to the current dominant psychophysiological state. One consequence of this variability in the representation of human states is that if the objective data used to train, test, and validate state prediction models is not recorded for a sufficiently long period, these records may not consistently capture the full spectrum of manifestations representing underlying psychophysiological states. Simultaneously, continuous monitoring requires a finer temporal resolution, which typically necessitates making predictions at shorter time intervals compared to the true values ​​required to capture the physiological dynamics that truly represent the target state. Therefore, human state prediction models using datasets with short windows as the true values ​​may be inaccurate or exhibit inconsistent sensitivity to different manifestations of the same underlying human state (e.g., the model may be highly sensitive to a first subset of manifestations of the same psychophysiological state but insensitive to a second subset); while models using longer windows may be unsuitable for continuous detection and monitoring.

[0004] The fact that no psychophysiological state is entirely singular further complicates the detection of human states. For example, a state of high cognitive load can be subdivided into aspects related to learning, decision-making, reaction, reflection, frustration, and so on. Even if subjects are assumed to consistently experience the target state, any of these different aspects of a particular psychophysiological state can have variable effects on physiological processes and will not be uniformly distributed over time. Similarly, particularly in the presence of microsleep episodes, observable levels of drowsiness and physiological performance may fluctuate, posing additional challenges to state prediction models that rely on data derived from short physiological recordings; alternatively, if, for example, a subject falls asleep while driving, a longer window may not provide sufficient temporal resolution to take any action.

[0005] Furthermore, the collection and analysis of physiological data for state detection is limited in terms of efficiency and resource utilization. Recruiting participants and modeling conditions to elicit target states are resource-intensive tasks. The limited availability of large datasets for modeling the correlation between physiological performance and underlying psychophysiological states is a widely recognized bottleneck in the field of human state detection and an obstacle to training robust and highly accurate human state prediction models.

[0006] Given these challenges, there is a need for improved methods to efficiently generate and utilize robust datasets for human state prediction that take into account the dynamics and variability of the physiological performance of potential human states. Such methods enable more accurate and efficient data collection, processing, and analysis, thereby facilitating the training and validation of continuous state prediction models and monitoring systems. This disclosure aims to address these and other related technical problems in the field of human state detection. Summary of the Invention

[0007] The current disclosure at least partially addresses the aforementioned problems. In one aspect, a method is provided for recording and analyzing physiological data to assess a subject's target state. The method includes recording the subject's physiological data while the subject remains in the target state for a continuous duration equal to or greater than a threshold duration sufficiently representing the target state in a selected biological signal for recording. The recorded physiological data is then segmented into a plurality of shorter analysis windows, each having a predetermined duration less than the threshold duration, and the plurality of shorter analysis windows are labeled with real value labels indicating the target state as a step toward sustained state detection. The method further includes storing the plurality of shorter analysis windows and associated real value labels in a non-transitory memory.

[0008] In another aspect, a system for initiating and evaluating a subject's target state is disclosed. The system includes a processor and a non-transitory memory storing instructions that, when executed by the processor, cause the system to initiate the subject's target state. The system records the subject's physiological data for a continuous duration equal to or greater than a threshold duration. The recorded physiological data is segmented into multiple shorter analysis windows, each with a predetermined duration equal to or less than the threshold duration. The system labels each of the multiple shorter analysis windows with a true value label indicating the target state and stores the multiple shorter analysis windows and the true value label in the non-transitory memory.

[0009] In another aspect, a method for training a state prediction model is provided. The method includes eliciting a target state in a subject and recording the subject's physiological data for a continuous duration greater than or equal to a threshold duration. The recorded physiological data is segmented into multiple physiological data windows, each with a predetermined duration less than the threshold duration. Each of the multiple physiological data windows is labeled with a ground truth label indicating the target state. The method includes storing the multiple physiological data windows and ground truth labels in a non-transitory memory. A training data pair is selected, comprising multiple physiological data windows with predetermined durations and ground truth labels indicating the target state. The method involves mapping the multiple physiological data windows to corresponding multiple state predictions using a state prediction model. A loss for the multiple state predictions is determined based on a loss function and the ground truth labels. The parameters of the state prediction model are updated based on the determined loss.

[0010] By utilizing long-duration physiological recordings to establish more typical true values ​​of human states, the disclosed method and system at least partially address the aforementioned challenges. This approach acknowledges the dynamic and variable nature of physiological performance, which may not be adequately captured by shorter, routine recordings. By recording physiological data over a continuous duration of at least a threshold duration, the method and system capture a broader range of physiological responses, thereby mitigating the risk of inaccuracies due to the sporadic and intermittent nature of human state expression. Subsequent segmentation of these extended recordings into shorter analytical windows (each smaller than the threshold duration) allows for continuous monitoring of the subject's state with enhanced temporal resolution. This segmentation not only reflects the variability of states such as cognitive load and drowsiness but also meets the need for high-frequency state detection in some applications, including driver alertness monitoring. Furthermore, the disclosed implementation significantly improves the efficiency of data collection and processing. By generating a larger dataset from long-duration recordings (which can be subdivided into multiple shorter windows), the method and system improve resource utilization efficiency. The expanded datasets achieved by the current disclosure provide a rich foundation for the application of machine learning and artificial intelligence algorithms, facilitating the development of more robust and accurate real-time human state detection models. Attached Figure Description

[0011] A better understanding of the various aspects of this disclosure can be achieved by reading the following detailed description and referring to the accompanying drawings, in which:

[0012] Figure 1 This is a schematic diagram of the process used to train a human state prediction model;

[0013] Figure 2 This is a block diagram of a human state prediction model training system;

[0014] Figure 3 This is a flowchart of a method for generating training data for a human state prediction model;

[0015] Figure 4 This is a flowchart of a method for detecting and labeling state behavior;

[0016] Figure 5 It is a graph comparing the physiological performance under low and high cognitive load conditions and high cognitive load conditions;

[0017] Figure 6 This is a flowchart of a method for training a human state prediction model;

[0018] Figure 7 This is a flowchart depicting a method for evaluating the discriminative power of a trained human state prediction model by comparing state predictions across different evoked states.

[0019] Figure 8These are graphical representations of two different methods for comparing human body states; and

[0020] Figure 9 This is a flowchart of a method for determining aggregate performance metrics for mathematical state prediction models. Detailed Implementation

[0021] This disclosure relates to methods and systems for generating and utilizing physiological data for human state detection, particularly for continuously monitoring human state within selected time windows. The methods and systems disclosed herein induce a target psychophysiological state in a subject, record physiological data for a continuous duration sufficient to capture a representation of the target state in recorded biosignals, segment the recorded data into shorter analysis windows as a step toward continuous state detection, label each window with a true value indicating the target state, and store the windows and labels in a non-transitory memory.

[0022] In one implementation scheme Figure 1 The document describes a process 100 for generating ground truth data and training a state prediction model. Process 100 captures and utilizes long-term physiological recordings to establish ground truth values ​​for various human states (such as cognitive load and drowsiness), which can be used for mathematical modeling and training machine learning (ML) and artificial intelligence (AI) models to achieve accurate state prediction. Process 100 can be performed by, for example... Figure 2 The human state prediction model training system 200, further detailed in the text, is implemented through... Figure 3 Method 300 is performed by one or more operations shown in the diagram. Method 300 includes collecting and labeling physiological data over a long period of time and processing such data into a shorter analysis window, which ensures that the training data covers a wide range of human state performance, thereby enhancing the model's ability to detect human states with greater accuracy.

[0023] Figure 4 A flowchart of a method 400 for detecting and labeling human state performance is shown. This method can be used for model training or for screening analysis windows prior to training, as well as for testing, validating, and comparing models. The method involves identifying specific physiological markers associated with different states, such as cognitive load or drowsiness, during various activities, including driving simulations. Figure 5 Graphical illustrations comparing physiological measures under low and high cognitive load conditions are provided. Figure 5 The illustrations highlight the diversity in physiological metrics / performance even when subjects consistently experience a single physical state.

[0024] Figure 6The training of a human state prediction model is depicted. The figure presents a flowchart of an implementation method 600 for training an ML model to predict the human state of a subject based on one or more physiological data windows. Figure 7 A flowchart is provided depicting a method 700 for evaluating the discriminative power of a trained state prediction model. This evaluation method compares state predictions across different triggering states to assess the model's accuracy and sensitivity to state changes. Figure 8 Graphical illustrations are provided for two different methods of comparing human states, such as those used in method 700 to evaluate the ability of a trained state prediction model to distinguish between a first and a second human state. Similarly, [the following methods can be used]. Figure 9 The method 900 shown determines the aggregate performance metrics of a mathematical human state prediction model (which may include a machine learning model in some implementations).

[0025] In summary, the disclosed system and method provide a novel approach for training more robust and accurate predictive models for human condition detection and continuous monitoring using physiological data over longer periods. This approach is significant for improving the accuracy and performance of continuous condition monitoring systems, particularly in scenarios such as vehicle operation or high-pressure working environments.

[0026] refer to Figure 1 The paper describes a human state prediction training process 100, illustrating an implementation scheme for generating physiological data and utilizing such data to develop a robust human state prediction model. Process 100 may be particularly advantageous in applications such as continuous monitoring of human state, where accurate prediction of human state based on physiological data is required.

[0027] Subject 102 refers to an individual from whom physiological data is acquired during process 100. In one implementation, particularly in scenarios where continuous monitoring is required, such as during a driving simulation, subject 102 may be a participant in a study designed to elicit a target human state (such as cognitive load or drowsiness). Subject 102 may be exposed to various tasks or conditions designed to induce the target state, and physiological responses may be recorded over a continuous duration.

[0028] Physiological data acquisition device 104 is configured to record physiological data of subject 102. In one embodiment, physiological data acquisition device 104 may include a combination of sensors and instruments, such as an electrocardiogram (ECG) monitor, an electroencephalogram (EEG) system, a photoplethysmography (PPG) sensor, a skin conductance sensor, and an eye-tracking device. These devices are used to capture a comprehensive set of physiological parameters that indicate the human state of subject 102 over a continuous duration of at least a certain threshold duration (e.g., five minutes).

[0029] Physiological data 106 includes one or more measures of physiological parameters recorded for a duration greater than a threshold duration (e.g., 5 minutes). While the duration is significantly longer than a few seconds in this example, the threshold duration can be adjusted based on indications of the subject's physiological parameters, such as respiratory rate, activity level, etc. To transform physiological data 106 into a more usable dataset, the data is segmented into multiple shorter analysis windows, each with a predetermined duration shorter than the first threshold duration (e.g., 30 seconds; although 30 seconds can be used as an example of the predetermined durations of multiple shorter analysis windows herein, it should be understood that durations greater or shorter than 30 seconds can be used, as long as the duration of the shorter analysis window is less than the threshold duration of the recorded physiological data). This segmentation allows for continuous monitoring of the subject 102's state with enhanced temporal resolution. In one embodiment, segmentation may involve overlapping consecutive analysis windows with a predetermined step size duration smaller than the predetermined duration of the analysis window, thereby ensuring a comprehensive representation of the body's state changing over time.

[0030] Human state prediction model 120 is a computational model designed to predict the human state of subject 102 based on physiological data 106. In one implementation, human state prediction model 120 may utilize an ML or AI algorithm trained using physiological data 106. Model 120 may be configured to map multiple physiological data windows to corresponding human state predictions, such as a first prediction 130, a second prediction 132, and an Nth prediction 134, where N is a positive integer greater than two.

[0031] First prediction 130, second prediction 132, and Nth prediction 134 represent a series of human state predictions generated by human state prediction model 120 for each of the shorter analysis windows derived from physiological data 106. In various embodiments, these predictions may be binary, categorical, or probabilistic in nature, reflecting the likelihood or existence of a target human state within each analysis window.

[0032] Loss function 140 is a mathematical structure used to quantify the error or loss between the predictions generated by human state prediction model 120 and the true value labels 150. In one implementation, loss function 140 can be a mean squared error function, a cross-entropy loss function, or any other suitable loss function known in the field of machine learning. Loss function 140 enables the training and optimization of human state prediction model 120 by determining the adjustments made to the model parameters during the training process.

[0033] The true value label 150 is assigned to the entire duration of the physiological data 106, and more broadly, to each of several shorter analysis windows. In one implementation, the true value label 150 may indicate the target human state that the subject 102 wants to experience during the data acquisition phase. The true value label 150 is used as a reference for comparison with the predictions of the human state prediction model 120.

[0034] Loss 160 is determined based on the difference between the predictions (first prediction 130 to Nth prediction 134) of physiological data 106 and a single true value label 150. In one implementation, loss 160 can be calculated based on the difference between the average prediction (made across the true value timescale) and the true value label 150, rather than comparing each prediction directly to the true value label 150. This method acknowledges the dynamic and variable nature of physiological performance in physiological metrics and aims to provide a more accurate assessment of the performance of the human state prediction model 120.

[0035] In an alternative implementation, the human state prediction training data generation process 100 may include additional components or operations, such as filtering multiple shorter analysis windows based on secondary ground truth labels that indicate the performance of the target human state, or optimizing the temporal resolution and accuracy of human state prediction by utilizing different window sizes and step durations.

[0036] In summary, the components and operations of the Human State Prediction Training Process 100 provide a comprehensive approach to generating training data for developing accurate and robust models for real-time human state detection.

[0037] refer to Figure 2 A human state prediction model training system 200 is shown for training models to predict human states based on physiological data. System 200 is configured to enhance the capabilities of continuous monitoring applications by providing accurate, relevant, and context-appropriate predictions of human states (such as cognitive load and drowsiness) using physiological data. System 200 includes a state prediction model training device 202, which is designed to interact with the various components and display the results via a display device 230.

[0038] The state prediction model training device 202 includes a processor 204 configured to execute machine-readable instructions stored in non-transitory memory 206. The processor 204 may be single-core or multi-core, and the program executing on it may be configured for parallel or distributed processing. In some embodiments, the processor 204 may include various components distributed across two or more devices that may be remotely located and / or configured for coordinated processing. In some embodiments, aspects of the processor 204 may be virtualized and executed by a remotely accessible networked computing device configured for cloud computing.

[0039] Non-transitory memory 206 stores machine-readable instructions that, when executed by processor 204, enable device 202 to perform various functions related to training the human state prediction model. Within non-transitory memory 206, a physiological data segmentation module 208 is stored. Physiological data segmentation module 208 is trained to segment physiological data into multiple shorter analysis windows, each with a predetermined duration less than a threshold duration (e.g., five minutes). In one embodiment, physiological data segmentation module 208 can utilize an algorithm to segment data with overlapping consecutive analysis windows at predetermined step lengths less than the predetermined duration of the analysis window, thereby expanding the amount of data and the diversity of possible data representations used for training and testing the persistent state prediction model.

[0040] The non-transitory memory 206 also stores training data 210. The training data 210 stores multiple physiological data windows, each uniquely associated with a corresponding true value label indicating the state of the target human body. In one embodiment, the training data 210 can be used to train one or more state prediction models 212 by performing one or more operations of method 600.

[0041] The state prediction model 212, also stored in non-transitory memory 206, is configured to map multiple physiological data windows to corresponding multiple human state predictions. In some embodiments, the state prediction model 212 may be an ML model configured to generate human state predictions based on physiological data such as ECG data, EEG data, PPG data, skin conductance data, and eye movement data.

[0042] A state performance detection module 214 is included within non-transitory memory 206. The state performance detection module 214 is configured to detect the occurrence of one or more of a predetermined set of performances of a target human state in each of a plurality of shorter analysis windows. In one embodiment, the state performance detection module 214 may employ machine learning techniques to identify specific physiological markers associated with different states (such as cognitive load or drowsiness) during various activities (including driving simulations).

[0043] User input device 250 is configured to interface with human state prediction model training device 202. User input device 250 may be a computer, smartphone, tablet, or any other device capable of submitting user input to system 200 and receiving responses. User input device 250 may include a user interface that allows users to interact with system 200, enter commands, and view results generated by state prediction model 212 based on data retrieved from training data 210.

[0044] Display device 230 is communicatively coupled to processor 204 and configured to display results and information related to the training of the human state prediction model. Display device 230 may include one or more display devices utilizing virtually any type of technology, such as a computer monitor, touchscreen, or projector. In some embodiments, display device 230 may be combined with processor 204 and non-transitory memory 206 in a shared housing, or it may be a peripheral display device.

[0045] Physiological data acquisition device 240 is configured to capture physiological data from a subject for a continuous duration of at least a threshold duration (e.g., five minutes). Physiological data acquisition device 240 may include various sensors and measurement tools, such as an ECG monitor, an EEG head-mounted device, a PPG sensor, a skin conductance sensor, and an eye-tracking device. In one embodiment, physiological data acquisition device 240 may be configured to record data in a simulated driving task environment to elicit a target human state in the subject. In another embodiment, device 240 may be adapted to capture physiological data in various settings, including clinical settings or during the performance of a cognitive task.

[0046] In an alternative implementation, components of the state prediction model training device 202 may include additional modules or features to enhance the system's capabilities. For example, system 200 may be adapted to support a variety of human states, making it suitable for a range of applications, from driver alertness monitoring to stress and fatigue detection in high-stress work environments.

[0047] refer to Figure 3 The diagram illustrates a flowchart of a method 300 for generating human training data. Method 300 can be adopted by a system (such as a human state prediction model training system 200) to train a continuous human state prediction model with enhanced accuracy and robustness.

[0048] At operation 302, the system induces a target human state in the subject. The induction of the target state can be achieved through various experimental devices designed to elicit specific human responses. In one embodiment, the system administers a cognitive task to the subject, such as an n-back task, a simulated driving task, or a pattern recognition task, to induce the target human state. In another embodiment, the system can modulate the cognitive task to adjust the level of cognitive load experienced by the subject. This modulation may involve altering the complexity of the cognitive task, the frequency of task stimuli, and the duration for which the subject performs the cognitive task. In a further embodiment, the system may employ a combination of sensory stimuli (such as auditory or visual cues) to induce the target human state, thereby simulating real-world conditions that may affect the subject's state, such as driving or operating a machine.

[0049] Proceeding to operation 304, the system records the subject's physiological data for a continuous duration of at least a threshold duration (five minutes). Recording physiological data for this extended duration allows for capturing a more complete range of the target human condition. In one embodiment, the system utilizes a suite of physiological data acquisition devices configured to measure heart rate, skin conductance, brain activity, and respiratory rate. In another embodiment, the physiological data includes at least one of ECG data, EEG data, PPG data, skin conductance data, and eye-tracking data. In a further embodiment, the system may employ wearable sensors that allow for the inconspicuous collection of physiological data while the subject performs cognitive tasks, thereby minimizing any potential interference with the subject's natural bodily responses.

[0050] At operation 306, the system segments the recorded physiological data into multiple shorter analysis windows. The predetermined duration of each analysis window is less than a threshold duration, allowing for continuous monitoring of the subject's state with enhanced temporal resolution. In one embodiment, the predetermined duration of the multiple shorter analysis windows is 30 seconds, which meets the needs of high-frequency state detection in some applications. In another embodiment, the system segments the recorded physiological data into multiple shorter analysis windows by overlapping consecutive analysis windows by a predetermined step duration less than the predetermined duration of the analysis window. This overlap ensures a more comprehensive capture of the variability in human state. In a further embodiment, the system may utilize advanced signal processing techniques to ensure that the segmentation of physiological data preserves the integrity of physiological signals, thereby maintaining data quality for subsequent analysis.

[0051] Moving to operation 308, the system is tasked with assigning a ground truth label to each of the segmented, shorter analysis windows, where each label reflects the target human state induced in the subject at operation 302. In one implementation, the ground truth label for each window is derived directly from the induced target human state without requiring complex reasoning or assumptions. This direct derivation ensures that the labels accurately represent the human state the subject intended to experience during the data acquisition phase. The labeling process establishes a reliable reference against which the performance of the human state prediction model can be evaluated.

[0052] At operation 310, the system uses secondary real values ​​to label each window based on the detected human state performance, such as... Figure 4 This is further detailed below. The operation involves identifying specific physiological markers associated with different states. In one implementation, the system detects the occurrence of one or more of a predetermined set of manifestations of a target human state within each of a plurality of shorter analysis windows. In another implementation, the system labels each of the plurality of shorter analysis windows with secondary ground truth labels indicating the detected manifestations of the target human state. In a further implementation, the system can refine the dataset used for model training by filtering the plurality of shorter analysis windows based on multiple corresponding secondary ground truth labels indicating the manifestations of the target human state.

[0053] Finally, at operation 312, the system stores the analysis window and the true value label in non-transitory memory. Storing the labeled analysis window in non-transitory memory facilitates subsequent retrieval and utilization of the data to train and validate the human state prediction model. In one embodiment, the system organizes the stored data in a structured database, thereby allowing efficient querying and access to the analysis window and associated labels. In another embodiment, the system may employ data encryption and access control mechanisms to ensure the privacy and security of the stored physiological data. After operation 312, method 300 can end.

[0054] In this way, method 300 can generate physiological data representing the dynamics and variability of human body states. By utilizing long-term physiological recordings, the system provides a rich dataset for the application, testing, validation, and comparison of machine learning and artificial intelligence algorithms, ultimately leading to the development of more robust and accurate real-time human state detection models.

[0055] refer to Figure 4The diagram illustrates a flowchart of a method 400 for detecting the performance of a target human state in a continuous recording of physiological data. Method 400 is able to identify and label human state performance within an analysis window derived from the continuous recording, thereby facilitating the establishment of ground truth values ​​for training and validating human state prediction models. Method 400 utilizes a series of operations to process physiological data, detect state performance, assign secondary ground truth labels, and store the processed data for subsequent use.

[0056] At operation 402, method 400 begins by receiving multiple analysis windows derived from continuous recording of physiological data. These analysis windows are segments of physiological data previously recorded for a continuous duration of at least a threshold duration. In one embodiment, the physiological data may include, but is not limited to, ECG data, EEG data, PPG data, skin conductance data, and eye-tracking data. Each analysis window has a predetermined duration, such as 30 seconds, and is designed to capture the dynamics and variability of the physiological manifestations of the target human state. In another embodiment, the analysis windows may be overlapped for predetermined step-length durations to ensure a comprehensive representation of the human state changing over time.

[0057] Proceeding to operation 404, the method involves initializing human state performance detection parameters. These parameters are used to detect state performance within each analysis window. In one embodiment, initialization may involve setting thresholds for physiological signal changes, patterns indicating the target state, and other criteria based on the type of physiological data being analyzed. For example, parameters for ECG data might include a heart rate variability threshold, while parameters for EEG data might focus on specific brainwave patterns associated with cognitive load or drowsiness. In another embodiment, initialization of the detection parameters may involve calibrating the system based on a training dataset that includes labeled examples of target human state performance.

[0058] At operation 406, the method includes detecting the performance of the target human state in each window. In one embodiment, the detection may involve applying a trained machine learning or artificial intelligence algorithm to identify patterns in physiological data that are associated with the target state. For example, a machine learning model may analyze ECG and EEG data to detect signs of cognitive load during a simulated driving task. In another embodiment, the detection may involve statistical analysis of physiological signals to identify deviations from baseline measures indicative of the target state, such as increased skin conductance in response to stress.

[0059] After detecting a state performance, operation 408 involves labeling each window with a secondary truth value label based on the detected performance. In one embodiment, the label may be binary, indicating the presence or absence of a specific performance from a set of predetermined performances of the target human state within the window. In another embodiment, the label may be categorical, reflecting the different levels or intensities of each of the multiple performances in the set of predetermined performances.

[0060] At operation 410, method 400 optionally filters multiple analysis windows based on the secondary ground truth labels associated with each window. This filtering process ensures that only the most typical windows are retained for further analysis and model training. In one implementation, windows with labels indicating the existence of the target state with high confidence can be selected, while those with lower confidence can be excluded. In another implementation, filtering may involve selecting windows that exhibit a range of representations of the target state to ensure that the predictive model can generalize to different representations of the state.

[0061] Finally, at operation 412, the method stores multiple analysis windows and associated secondary ground truth labels in a non-transitory memory. This storage facilitates access to and retrieval of processed data for future use in training and validating human state prediction models. In one embodiment, the data can be organized in a structured database, allowing for efficient querying based on various criteria, such as the type of physiological data or the intensity of the target state. In another embodiment, the data can be encrypted and access controlled to ensure the privacy and security of the information. After operation 412, method 400 can terminate.

[0062] refer to Figure 5 This figure depicts a comparison of physiological data under different cognitive load conditions. The graphical representation provides a visual comparison between physiological data under low cognitive load 500A and high cognitive load 500B. This comparison illustrates the dynamics and variability of physiological processes associated with cognitive load, and how different human states may depend on a small number of underlying physiological manifestations that are randomly distributed over time.

[0063] The physiological data under low cognitive load 500A are characterized by a set of physiological measures 502A, which may include, but are not limited to, heart rate, ECG data, EEG data, PPG data, skin conductance data, and eye movement data. These measures indicate the subject's physical state under minimal cognitive demands. In one embodiment, physiological measures 502A may represent a baseline or control condition, where the subject is at rest or engaged in a task requiring minimal cognitive effort. In another embodiment, physiological measures 502A may be obtained during passive observation or during simple repetitive tasks that do not significantly consume cognitive resources.

[0064] The first window 504A through the Nth window 506A represent discrete segments of continuous physiological data recorded within a threshold duration (such as five minutes or longer). While the remainder of the description illustrates example operation using a 5-minute threshold as the threshold duration, it should be understood that other threshold durations can be used, as explained herein. Each window (such as the first window 504A) has a predetermined duration, e.g., 30 seconds, and is labeled with a true value indicating a low cognitive load state. The Nth window 506A represents the last of a series of such windows. In one embodiment, the continuous analysis windows may overlap for a predetermined step length duration, such as one second, to ensure a comprehensive representation of the physiological data. In another embodiment, the windows may not overlap, thus providing different snapshots of the physiological state at different time intervals.

[0065] Physiological data under high cognitive load 500B are similarly characterized to physiological measure 502B, which is recorded when subjects are subjected to tasks designed to elicit high cognitive load. These tasks may include, but are not limited to, n-back tasks, simulated driving tasks with additional cognitive demands, or complex pattern recognition tasks. Physiological measure 502B captures enhanced human responses associated with increased cognitive effort, such as changes in heart rate variability, EEG patterns indicating emphasized mental engagement, or changes in blink and scan rates.

[0066] The first high cognitive load manifestation 520 and the second high cognitive load manifestation 522 represent specific features or patterns within physiological data that indicate high cognitive load. For example, the first high cognitive load manifestation 520 may correspond to a specific heart rate variability pattern known to be associated with intense cognitive processing. The second high cognitive load manifestation 522 may represent different EEG waveform patterns, such as increased theta wave activity, which is typically associated with tasks requiring significant attention and working memory resources. These manifestations are detected within each of shorter analysis windows (such as the first window 504B and the Nth window 506B) and are labeled with secondary true value labels indicating the detected manifestations of a high cognitive load state.

[0067] The first window 504B and the Nth window 506B under high cognitive load conditions are similar to those described under low cognitive load conditions, where each window represents a segment of continuous physiological data. However, these windows are expected to show changes in physiological measures that reflect the increased cognitive demands imposed on the subject. In one embodiment, because the subject's physical state fluctuates in response to the cognitive task, the windows under high cognitive load conditions may show greater changes in physiological measures compared to low cognitive load conditions. In another embodiment, the windows may reveal patterns of physiological response associated with specific phases or subprocesses involved in the cognitive task, such as the onset of a decision-making process or a period of reflection following a task response.

[0068] refer to Figure 6 The diagram illustrates a flowchart of a method 600 for training a human state prediction model using physiological data. Method 600 is configured to enhance the capabilities of continuous monitoring applications by leveraging physiological data to provide accurate predictions of human states such as cognitive load and drowsiness.

[0069] At operation 602, the system selects a training data pair, which includes multiple physiological data windows and associated ground truth labels indicating the target human state. In one implementation, the physiological data windows can be derived from continuous recordings of physiological data, such as electrocardiogram (ECG), EEG, PPG, skin conductance, and eye movement data recorded over a duration of at least five minutes. The ground truth labels are assigned based on the target human state the subject intends to experience during the data acquisition phase. In another implementation, the training data pairs can be filtered based on secondary ground truth labels associated with detected performance of the target human state, thereby ensuring that the model is trained on data that accurately represents the variable performance of the state.

[0070] Continuing to operation 604, the system uses a human state prediction model to map multiple physiological data windows to corresponding human state predictions. This mapping process involves applying the prediction model to each physiological data window to generate a prediction of the human state. In one implementation, the prediction model can be an ML model that uses features extracted from the physiological data (such as heart rate variability, brain wave patterns, or skin conductance levels) to predict the state. The model can employ algorithms such as support vector machines, neural networks, or decision trees to perform the mapping. In another implementation, the system can utilize AI techniques (such as deep learning) to map complex patterns within the physiological data to human state predictions, thereby enabling the detection of subtle manifestations of the target state.

[0071] At operation 606, the system determines the loss for predicting multiple human states based on a loss function and the ground truth labels. The loss function quantifies the error or difference between the predictions generated by the model and the ground truth labels. In one implementation, the loss function may be a mean squared error function, which measures the average squared difference between the predicted values ​​and the ground truth human labels. In another implementation, the loss function may be a cross-entropy loss function, which is particularly suitable for classification tasks where the output is a probability distribution about different states.

[0072] Switching to Operation 608, the system updates the parameters of the human state prediction model based on a defined loss. Updating the model parameters is an iterative process involving adjusting the weights and biases within the model to reduce the loss. In one implementation, the system may employ optimization algorithms (such as gradient descent or stochastic gradient descent) to perform the updates. These algorithms calculate the gradient of the loss function with respect to the model parameters and adjust in the direction that minimizes the loss. In another implementation, the system may utilize regularization techniques, such as L1 or L2 regularization, to prevent overfitting and ensure that the model is adequately generalized to new data.

[0073] At operation 610, the system determines the human state discrimination capability of the human state prediction model. This determination involves evaluating the model's ability to distinguish different human states. In one implementation, the system compares predictions generated for a first plurality of physiological data windows acquired when a subject is induced to a first human state with predictions generated for a second plurality of physiological data windows acquired when a subject is induced to a second human state. This comparison evaluates the model's sensitivity and specificity in distinguishing different states. In another implementation, the system can validate an updated human state prediction model by comparing a set of human state predictions with a separate validation set of physiological data windows and associated true value labels. The validation set can be different continuously recorded sessions of physiological data from the subject or different subjects, thereby ensuring the model's discrimination capability is robust across different subjects and conditions.

[0074] By utilizing long-duration physiological recordings as ground truth, Method 600 addresses the dynamic and variable nature of physiological manifestations, which pose challenges to the accurate detection of human states such as cognitive load and drowsiness. This method systematically segments these recordings into shorter analytical windows (each labeled with a ground truth indicating the target human state), enabling the development of models sensitive to subtle expressions of human states and robust across variable manifestations. Furthermore, Method 600 evaluates the model's discriminative power by comparing predictions across different eliciting states to ensure the model can distinguish between various human states, thereby enhancing its applicability in real-time monitoring scenarios. Therefore, Method 600 represents a significant advancement in the continuous monitoring of human states, providing a foundation for developing accurate and reliable human state prediction models that can be advantageously utilized in applications requiring high-frequency state detection and analysis.

[0075] refer to Figure 7 The diagram illustrates a flowchart of a method 700 for evaluating the discriminative power of a human state prediction model. Method 700 utilizes a series of operations to process physiological data, generate state predictions, and evaluate the model's ability to distinguish various human states. This can be beneficial in developing persistent state prediction models, for example, for comparing prediction models trained with different hyperparameters or training data.

[0076] At operation 702, the system uses a human state prediction model to map each of a first plurality of physiological data windows acquired when inducing a subject's first human state to a first plurality of human state predictions. The first human state can be one of several states, such as cognitive overload or drowsiness, induced in the subject by various experimental devices designed to elicit specific human responses. In one embodiment, the system administers a cognitive task to the subject, such as an n-back task, a simulated driving task, or a pattern recognition task, to induce the first human state. A physiological data window is a segment of physiological data previously recorded over a continuous duration of at least five minutes. In another embodiment, the physiological data includes at least one of ECG data, EEG data, PPG data, skin conductance data, and eye movement data. Each analysis window has a predetermined duration, such as 30 seconds, and is designed to capture the dynamics and variability of the physiological manifestations of the target human state.

[0077] Continuing to operation 704, the system uses a human state prediction model to map each of the second plurality of physiological data windows acquired when inducing the subject's second human state to a second plurality of human state predictions. The second human state differs from the first human state and may represent different levels of cognitive load, different types of stress, or different stages of drowsiness, or controls in the absence of the first human state (e.g., ensuring the subject is not drowsy or under low cognitive load). In one embodiment, the system may adjust the cognitive task to induce the second human state, changing the complexity of the task, the frequency of task stimuli, or the duration for which the subject performs the task. In another embodiment, the system may employ a combination of sensory stimuli (such as auditory or visual cues) to induce the second human state, thereby simulating real-world conditions that may affect the subject's state, such as driving or operating a machine.

[0078] At operation 706, the system determines the human state discrimination capability of the human state prediction model based on a first plurality of human state predictions and a second plurality of human state predictions. This determination involves evaluating the model's ability to distinguish between the first and second human states. In one implementation, the system compares the predictions generated for the first plurality of physiological data windows with the predictions generated for the second plurality of physiological data windows to evaluate the model's accuracy and sensitivity to state changes. As will be discussed below... Figure 8 The description discusses this in more detail, and the discriminative power of the human state prediction model can be determined according to the first embodiment, in which the average value of a first plurality of predictions generated for a first human state is determined and compared with the average value determined for a second plurality of predictions generated for a second human state, such as... Figure 8 The many-to-many comparison 800B shown is illustrated. In another embodiment, a first prediction generated for a first human state can be compared with each of a plurality of predictions generated for a second human state to produce an average difference between the first prediction and the plurality of predictions for the second human state. This process can be repeated for each of the first plurality of predictions generated for the first human state, thereby producing a one-to-many comparison for each of the first plurality of human state predictions, such as... Figure 8 The one-to-many comparison shown is depicted in 800A.

[0079] In this way, Method 700 evaluates the model’s discriminative ability by comparing predictions across different eliminative states, so as to ensure that the model can distinguish between various human states, thereby enhancing its applicability in real-time monitoring scenarios.

[0080] refer to Figure 8The figure illustrates a graphical representation of human body data comparisons, showcasing one-to-many and many-to-many comparison methods used to evaluate the discriminative power of human body state prediction models. The figure is divided into two main sections: one-to-many comparison 800A and many-to-many comparison 800B, each demonstrating a method for comparing human body state predictions to assess the model's accuracy and sensitivity to state changes.

[0081] In a one-to-many comparison 800A, the first human state 820 is represented by a series of windows, including windows 804, 806, and 808. Each window corresponds to a shorter analysis window derived from continuous recording of physiological data, as described herein. These windows capture the dynamics and variability of physiological manifestations associated with the first human state 820, which may include states such as cognitive load or drowsiness during activities such as driving.

[0082] In one implementation, window 804 may represent a 30-second segment of physiological data in which the subject experiences high cognitive load due to a complex driving simulation task. Windows 806 and 808 may represent subsequent 30-second segments, capturing different aspects of the high cognitive load state, such as decision-making and reaction reflection. These windows are analyzed to generate a prediction of the human state, which is then compared with a second human state 840 (represented by windows 810, 812, and 814). The second human state 840 may reflect different levels of cognitive load or a control state with minimal cognitive demands.

[0083] In another implementation, a one-to-many comparison 800A may involve comparing a first prediction generated for a first human state 820 with each of a plurality of predictions generated for a second human state 840. This comparative evaluation model is based on the ability of physiological data captured in the corresponding window to distinguish between the two states.

[0084] Many-to-many comparison 800B extends the comparison method by combining the predicted averages of the first and second human states. The first average 850 represents the average of the predictions from windows 804, 806, and 808, which collectively characterize the first human state 820. This average serves as a typical measure of the model's predictions for the first state, capturing the overall trend of physiological performance within a selected time range.

[0085] In one implementation, a first average 850 can be calculated by averaging the output probabilities or classification predictions generated by the human state prediction model within each window associated with the first human state 820. This average provides a comprehensive view of the model's performance across multiple segments of physiological data, reflecting the combined effect of various performances of the target state.

[0086] On the other hand, the second average 860 represents the average of the predictions from windows 810, 812, and 814, which collectively characterize the second human state 840. In a similar implementation, the second average 860 can be determined by averaging the model predictions within each window associated with the second state, thereby providing a metric comparable to the first average 850.

[0087] In another implementation, the many-to-many comparison 800B may involve calculating the average difference between a first average 850 and a second average 860 to assess the discriminative power of the human state prediction model. This comparison can reveal the model's ability to distinguish between a first human state and a second human state based on aggregated predictions for each state. The comparison can also take into account the diversity within each set of windows, thus providing insights into the model's consistency and reliability in state predictions.

[0088] Figure 8 The comparisons described can be applied to a variety of human states beyond cognitive load and drowsiness, such as stress or fatigue, and can be used in various application areas, including automotive safety, healthcare monitoring, and workplace ergonomics.

[0089] refer to Figure 9 A flowchart of a method 900 for evaluating the performance of a mathematical model to predict physiological states based on continuous physiological data recordings is shown. Method 900 is capable of evaluating the accuracy and reliability of a mathematical model in the context of human state detection. Method 900 advantageously combines multiple predictions made within a shorter duration physiological data window into an aggregated performance metric that better represents the predictive ability of the mathematical model for continuous physiological data recordings, while still enabling the mathematical model to make human state predictions within a shorter duration physiological data window.

[0090] Method 900 begins with operation 902, in which the system selects multiple physiological data windows derived from continuous physiological data recording, along with corresponding true-value state labels. Continuous physiological data recording is acquired for a duration equal to or greater than a threshold duration to ensure broad physiological manifestations of the target state are captured. The predetermined duration of each physiological data window has a predetermined duration that is less than the threshold duration, thereby allowing for continuous monitoring of the subject's state with enhanced temporal resolution. In another implementation, the true-value state label indicates the target state, such as cognitive load or drowsiness that the subject is expected to experience during the data acquisition phase.

[0091] Continuing to operation 904, the system uses a mathematical model to predict the state of each of multiple physiological data windows. This mathematical model is configured to map the physiological data windows to corresponding state predictions. In one implementation, the mathematical model can be a machine learning model that uses features extracted from the physiological data to predict states. The model can employ algorithms such as support vector machines, neural networks, or decision trees to perform the predictions. In another implementation, the system can utilize artificial intelligence techniques (such as deep learning) to map complex patterns within the physiological data to state predictions, thereby enabling the detection of subtle manifestations of the target state.

[0092] In one implementation, the mathematical model can be a supervised machine learning model trained on a dataset that includes a window of physiological data and corresponding ground truth state labels. The model can use various features extracted from the physiological data, such as heart rate variability, brainwave patterns, or skin conductance levels, to predict its state. These features can be selected based on their correlation with the target state and their ability to distinguish between different states. For example, features strongly correlated with cognitive load can be used to predict states of high cognitive demand.

[0093] In another implementation, mathematical models that do not rely on machine learning or deep learning (such as statistical models or rule-based systems) can be used to predict states based on physiological data. These models can employ predefined rules, statistical analysis, or mathematical equations to infer the target state from the physiological data. For example, statistical models can use regression analysis to determine the relationship between physiological measurements and target states, while rule-based systems can apply a set of expert-defined rules to classify physiological data into different states.

[0094] At operation 906, the system uses the true value state labels to determine multiple performance metrics for the mathematical model, with one performance metric used for each state prediction. These performance metrics quantify the accuracy of the model's predictions compared to the true values. In one implementation, the performance metrics may include measures such as precision, recall, F1 score, and accuracy. In another implementation, the system may compute a confusion matrix to visualize the model's performance across different states, thereby providing insight into the model's sensitivity and specificity.

[0095] At operation 908, the system aggregates multiple performance metrics to produce an aggregated performance metric for continuous physiological data recording. This aggregation provides a comprehensive view of the overall model performance. In one implementation, the aggregated performance metric can be a weighted average of the individual performance metrics, taking into account the relative importance of each metric. In another implementation, the system can employ statistical methods to combine the performance metrics, such as calculating the area under the receiver operating characteristic (ROC) curve or a precision-recall curve, which indicates the model's discriminative ability.

[0096] At operation 910, the system stores aggregated performance metrics in non-transitory memory. Storing aggregated performance metrics facilitates subsequent retrieval and utilization of the data for further analysis or model refinement. In one implementation, the system organizes the stored data in a structured database, thereby allowing efficient querying and access to performance metrics and associated tags. After operation 910, method 900 can terminate.

[0097] Method 900 generates robust performance evaluations for mathematical models used for human state detection. Method 900 systematically merges predictions from segmented physiological data windows into an aggregated performance metric, thereby providing a quantitative assessment of the mathematical model's predictive accuracy across extended physiological recordings. This method ensures the determination of accurate performance metrics for the model while ensuring that the model retains its predictive power within shorter duration data windows, thus facilitating robust human state detection across different time scales.

[0098] This disclosure also provides support for a method comprising: recording physiological data of a subject in a target state for a continuous duration equal to or greater than a threshold duration; segmenting the recorded physiological data into a plurality of shorter analysis windows, each analysis window having a predetermined duration less than the threshold duration; labeling each of the plurality of shorter analysis windows with a true value label indicating the target state; and storing the plurality of shorter analysis windows and the true value label in a non-transitory memory. In a first example of the method, the method further comprises: inducing the target state of the subject by: managing one or more of a cognitive task, an audiovisual stimulus, and an environmental factor to induce the target state; and adjusting the level of the target state by changing the complexity of the cognitive task, adjusting the intensity of the audiovisual stimulus, or adjusting the intensity of the stress factor. In a second example of the method, which optionally includes the first example, the physiological data includes at least one of electrocardiogram (ECG) data, electroencephalogram (EEG) data, photoplethysmography (PPG) data, skin conductance data, and eye movement data. In a third example, which optionally includes one or both of the first and second examples, the method further includes: detecting the occurrence of one or more of a predetermined set of manifestations of the target state in each of the plurality of shorter analysis windows; and labeling each of the plurality of shorter analysis windows with a secondary real value label indicating the detected manifestation of the target state. In a fourth example, which optionally includes one or more or each of the first to third examples, the method further includes: filtering the plurality of shorter analysis windows based on a plurality of corresponding secondary real value labels indicating the manifestation of the target state. In a fifth example, which optionally includes one or more or each of the first to fourth examples, segmenting the recorded physiological data into the plurality of shorter analysis windows includes selecting a window size and dividing the recorded physiological data into consecutive analysis windows according to a predetermined step duration less than the predetermined duration of the plurality of shorter analysis windows. In a sixth example, which optionally includes one or more of the first to fifth examples, or each of them, the method further includes: predicting the state of each of the plurality of shorter analysis windows using a mathematical model, wherein the mathematical model is configured to map the plurality of shorter analysis windows to a plurality of state predictions. In a seventh example, which optionally includes one or more of the first to sixth examples, or each of them, the method further includes: determining a plurality of performance metrics for the mathematical model using the true value labels, each of the plurality of state predictions having a performance metric, wherein the plurality of performance metrics includes at least one of precision, recall, F1 score, and accuracy; and aggregating the plurality of performance metrics to produce an aggregated performance metric of the recorded physiological data, wherein the aggregated performance metric is a weighted average of the plurality of performance metrics.

[0099] This disclosure also provides support for a system for inducing and assessing a target state of a subject, the system comprising: a processor; and a non-transitory memory storing instructions that, when executed by the processor, cause the system to: induce the target state of the subject; record physiological data of the subject for a continuous duration equal to or greater than a threshold duration; segment the recorded physiological data into a plurality of shorter analysis windows, each analysis window having a predetermined duration equal to or less than the threshold duration; label each of the plurality of shorter analysis windows with a real value label indicating the target state; and store the plurality of shorter analysis windows and the real value label in the non-transitory memory. In a first example of the system, the processor is further configured to: manage a cognitive task for the subject to induce the target state, the cognitive task including one of an n-back task, a simulated driving task, and a pattern recognition task; and adjust the level of cognitive load experienced by the subject by varying the complexity of the cognitive task, the frequency of task stimuli, and the duration for which the subject performs the cognitive task. In a second example, which optionally includes the first example, the processor is further configured to: detect the occurrence of one or more of a predetermined set of manifestations of the target state in each of the plurality of shorter analysis windows; and label each of the plurality of shorter analysis windows with a secondary real value label indicating the detected manifestation of the target state. In a third example, which optionally includes one or both of the first and second examples, the processor is further configured to: filter the plurality of shorter analysis windows based on a plurality of corresponding secondary real value labels indicating the manifestation of the target state. In a fourth example, which optionally includes one or more or each of the first to third examples, the predetermined duration of the plurality of shorter analysis windows is 30 seconds. In a fifth example, which optionally includes one or more or each of the first to fourth examples, the processor is further configured to: segment the recorded physiological data into the plurality of shorter analysis windows by overlapping consecutive analysis windows with a predetermined step duration smaller than the predetermined duration of the plurality of shorter analysis windows.

[0100] This disclosure also provides support for a method for training a state prediction model, the method comprising: inducing a target state in a subject; recording physiological data of the subject for a continuous duration greater than or equal to a threshold duration; segmenting the recorded physiological data into a plurality of physiological data windows, each of the plurality of physiological data windows having a predetermined duration less than the threshold duration; labeling each of the plurality of physiological data windows with a real value label indicating the target state; storing the plurality of physiological data windows and the real value label in a non-transitory memory; selecting a training data pair comprising a plurality of physiological data windows having a predetermined duration and a real value label indicating the target state; mapping the plurality of physiological data windows to corresponding plurality of state predictions using the state prediction model; determining a loss for the plurality of state predictions based on a loss function and the real value label; and updating parameters of the state prediction model based on the determined loss. In a first example of the method, selecting the training data pair further comprises filtering the plurality of physiological data windows based on a secondary real value label associated with a detected performance of the target state. In a second example, which optionally includes the first example, recording the subject's physiological data further includes utilizing at least one physiological data acquisition device configured to measure one or more of heart rate, skin conductance, brain activity, and respiratory rate. In a third example, which optionally includes one or both of the first and second examples, the method further includes determining the state discrimination capability of the state prediction model by comparing predictions generated for a first plurality of physiological data windows acquired when inducing a first human state in the subject with predictions generated for a second plurality of physiological data windows acquired when inducing a second human state in the subject. In a fourth example, which optionally includes one or more or each of the first to third examples, the method further includes validating the updated state prediction model by comparing the state prediction set with a separate validation set of the data windows and associated ground truth labels. In a fifth example, which optionally includes one or more or each of the first to fourth examples, the separate validation set of the physiological data windows is derived from different continuous data recording sessions of the subject or different subjects.

[0101] The foregoing description of aspects of this disclosure is based on flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a dedicated processor, or a field-programmable processor.

[0102] While the foregoing describes embodiments of this disclosure, other and additional embodiments of this disclosure may be devised without departing from the basic scope of this disclosure, the scope of which is defined by the following claims.

Claims

1. A method comprising: Record physiological data of subjects in the target state for a continuous duration equal to or greater than the threshold duration; The recorded physiological data is divided into multiple shorter analysis windows, each with a predetermined duration less than a threshold duration. Each of the plurality of shorter analysis windows is labeled with a true value label indicating the target state; as well as The multiple shorter analysis windows and the true value labels are stored in non-temporary memory.

2. The method of claim 1, further comprising: The target state of the subject is induced by the following steps: Managing one or more of cognitive tasks and audiovisual stimuli to elicit the target state; as well as The level of the target state is adjusted by changing the complexity of the cognitive task or by adjusting the intensity of the audiovisual stimuli.

3. The method of claim 1, wherein the physiological data includes at least one of electrocardiogram (ECG) data, electroencephalogram (EEG) data, photoplethysmography (PPG) data, skin conductance data, and eye movement data.

4. The method of claim 1, further comprising: In each of the plurality of shorter analysis windows, the occurrence of one or more of a predetermined set of manifestations of the target state is detected; as well as Each of the plurality of shorter analysis windows is labeled with a secondary true value label indicating the detected performance of the target state.

5. The method of claim 4, further comprising: The multiple shorter analysis windows are filtered based on multiple corresponding secondary truth value labels that indicate the performance of the target state.

6. The method of claim 1, wherein dividing the recorded physiological data into the plurality of shorter analysis windows includes selecting a window size and dividing the recorded physiological data into consecutive analysis windows according to a predetermined step duration smaller than the predetermined duration of the plurality of shorter analysis windows.

7. The method of claim 1, further comprising: A mathematical model is used to predict the state of each of the plurality of shorter analysis windows, wherein the mathematical model is configured to map the plurality of shorter analysis windows to a plurality of state predictions.

8. The method of claim 7, further comprising: The true value labels are used to determine multiple performance metrics of the mathematical model, each of the multiple state predictions having a performance metric, wherein the multiple performance metrics include at least one of precision, recall, F1 score and accuracy; as well as The aggregated performance metrics are used to generate an aggregated performance metric for the recorded physiological data, wherein the aggregated performance metric is a weighted average of the aggregated performance metrics.

9. A system for inducing and assessing a target state of a subject, the system comprising: processor; as well as A non-transitory memory that stores instructions, which, when executed by the processor, cause the system to: To induce the target state in the subject; The subject's physiological data were recorded for a continuous duration equal to or greater than the threshold duration. The recorded physiological data is divided into multiple shorter analysis windows, each with a predetermined duration equal to or less than a threshold duration. Each of the plurality of shorter analysis windows is labeled with a true value label indicating the target state; and The multiple shorter analysis windows and the true value labels are stored in the non-temporary memory.

10. The system of claim 9, wherein the processor is further configured to: The subject is given a cognitive task to elicit the target state, the cognitive task including one of an n-back task, a simulated driving task, and a pattern recognition task; and The level of cognitive load experienced by the subject was adjusted by changing the complexity of the cognitive task, the frequency of the task stimuli, and the duration of the subject's performance of the cognitive task.

11. The system of claim 9, wherein the processor is further configured to: In each of the plurality of shorter analysis windows, the occurrence of one or more of a predetermined set of manifestations of the target state is detected; and Each of the plurality of shorter analysis windows is labeled with a secondary true value label indicating the detected performance of the target state.

12. The system of claim 11, wherein the processor is further configured to: The multiple shorter analysis windows are filtered based on multiple corresponding secondary truth value labels that indicate the performance of the target state.

13. The system of claim 9, wherein the predetermined duration of the plurality of shorter analysis windows is 30 seconds.

14. The system of claim 9, wherein the processor is further configured to: The recorded physiological data is segmented into the plurality of shorter analysis windows by overlapping consecutive analysis windows with a predetermined step duration smaller than the predetermined duration of the plurality of shorter analysis windows.

Images