Systems and methods for detection of delirium and other neurological conditions - Patents.com

Abstract

Described herein are systems and methods for detecting and monitoring delirium in a subject. Other neurological conditions may also be detected and monitored. The system may include a data module configured to acquire a plurality of electroencephalography (EEG) signals collected from a subject. The system may also include a processing module in communication with the data module. The processing module may be configured to process the data and detect and monitor delirium and / or one or more other neurological conditions that the subject is suffering from or likely to be suffering from. The processing module may also generate an indication or assessment regarding delirium and / or for each neurological condition, at an individual level, or optionally between two or more concomitant neurological conditions.

Description

[Technical field]

[0001] (CROSS REFERENCE TO RELATED APPLICATIONS) This application claims priority to U.S. Provisional Application No. 63 / 298,937, filed January 12, 2022, which is incorporated by reference in its entirety herein.

[0002] The present application relates to a system and method for detecting and monitoring delirium in a subject. The detection and monitoring of delirium may be based on the output from one or more machine learning models that process a plurality of features extracted from electroencephalography (EEG) signals. The output of the one or more machine learning models may be used to provide or generate delirium propensity of a subject over a period of time. The system and method may also be employed to detect and monitor other neurological conditions or brain function abnormalities, such as seizures, strokes, and sedation, and to differentiate between conditions. [Background technology]

[0003] Delirium is a clinical condition that manifests as an acute disturbance in cognitive function and is common among hospitalized patients, especially in certain high-risk patient populations. The American Psychiatric Association's Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-V) details the following diagnostic criteria for delirium: a) impairment in attention (i.e., impaired ability to orient, focus, sustain, or shift attention) and consciousness (reduced orientation to the environment); b) impairment that is of short-term onset (usually hours to days) and that represents an acute alteration from baseline levels of attention and consciousness and that tends to fluctuate in severity over the course of a day; c) additional impairment in cognitive function (e.g., memory deficits, disorientation, language, visuospatial awareness, or perception); d) impairment in criteria a) and c) that is not better explained by a pre-existing, established, or ongoing neurocognitive disorder; and e) evidence from the history, physical examination, and laboratory findings that the impairment is caused by a direct physiological consequence of another medical illness, substance intoxication or withdrawal (i.e., from a drug of abuse or pharmacological therapy), or exposure to a toxic substance, or has multiple etiologies.

[0004] Delirium can exist in hypoactive or hyperactive states. Hyperactive delirium is characterized by increased activity and can include restlessness, agitation, or aggressive behavior. Hypoactive delirium is characterized by decreased activity and can include lethargy, abnormal sleepiness, or social withdrawal. Mixed delirium refers to fluctuations between hypoactive and hyperactive delirium states.

[0005] Failure to diagnose delirium has been shown to significantly impact patient mortality. Monitoring of delirium and other neurological conditions, such as sedation, is traditionally performed by trained health professionals, physicians, or nurses, using various assessment scales via clinical assessment. This practice can be problematic because assessment methods generally rely on physical patient movement or reaction, making subtle distinctions between various sedation and delirium levels difficult. For example, the current Standard of Care (SOC) for delirium assessment in acute care settings is the Confusion Assessment Matrix-ICU (CAM-ICU). Although the CAM-ICU was designed for use specifically in intensive care units, it has become the SOC for all acute care settings, including emergency departments and other critical care settings. The CAM-ICU assessment is typically performed by a bedside nurse and consists of a series of assessment features, such as: Feature 1: Acute onset or fluctuating course. o Has the patient had any fluctuations in mental status in the past 24 hours that differ from their baseline mental status or that are evidenced by fluctuations on the level of sedation / consciousness scale? ·Characteristic 2: Attention deficit. The assessor slowly reads out a series of 10 letters. The patient is asked to squeeze the assessor's hand each time the letter "A" is heard. This feature is failed if the patient makes more than two errors. · Feature 3: Altered level of consciousness. o The patient is in any state other than restless and calm? Using the Richmond Agitation-Sedation Scale (RASS), the score is any value other than 0? ·Characteristic 4: Disorganized thinking. The assessor asks a series of logic questions. This feature is failed if the patient gets it wrong more than once. -Do stones float on water? -Are there fish in the sea? -Is 1 pound heavier than 2 pounds? -Can you use a hammer to drive in a nail?

[0006] A patient is considered positive for delirium if feature 1, feature 2, and either feature 3 or 4 are present. The CAM-ICU assessment has been validated to be highly effective when performed within a research setting. However, in real-world routine use, it has become evident that the validity of the delirium assessment tool is significantly worse compared to the results obtained in a research setting.

[0007] The most comprehensive study of the validity of the CAM-ICU assessment in real-world routine use conditions included 282 patients across 10 different hospital ICUs. In this study, bedside clinical ICU nurses, all of whom were trained in the CAM-ICU, achieved a sensitivity of only 47% with a specificity of 98%. More than half of patients with delirium were not detected using the standard-of-care clinical nurse assessment in the 10 participating hospitals. Other smaller studies have also shown that in real-world routine settings, nurses may fail to recognize delirium 75% of the time.

[0008] There is also no SOC to assess delirium with sufficient frequency to adequately manage the fluctuating nature of delirium. The delirium assessment tools used in the current SOCs (including the CAM-ICU) were designed to be administered once per shift by the bedside clinical nurse. This means that, at most, delirium assessments are administered twice per day (once every 12 hours). In some hospitals, delirium assessments are only administered once per day during morning rounds by the medical team. As a result, SOC delirium assessments can suffer from delays in delirium recognition of 12 to 24 hours. Studies have shown that delayed treatment of delirium can result in significantly increased patient morbidity and mortality.

[0009] Delirium is known to be multifactorial, with many possible causes and risk factors. An important first step in managing patients with delirium is to recognize the risk factors and reduce or eliminate them where possible. Some of the delirium risk factors, such as age, are not modifiable. However, other risk factors, such as the use of delirium-inducing medications, are potentially modifiable. Treatment of delirium may consist of non-pharmacological as well as pharmacological interventions. Non-pharmacological interventions include sleep optimization, re-adaptation, increased exercise, family involvement, and other sensory / behavioral interventions. Pharmacological interventions can include stopping or altering existing medications that increase the risk of delirium, or administering medications that are intended to reduce or alleviate delirium.

[0010] Given the multifactorial nature of delirium and the many possible treatment options, monitoring treatment effectiveness may be an important part of optimizing treatment and minimizing delirium duration. However, the SOC delirium assessment method is unable to provide continuous monitoring due to the infrequent nature of the assessment. Therefore, it would be beneficial to have alternative methods and systems for detecting and monitoring delirium. It would also be beneficial to have new methods and systems for detecting and monitoring other neurological conditions. Summary of the Invention [Means for solving the problem]

[0011] Described herein are systems and methods that can rapidly and accurately use EEG signals and machine learning to detect and monitor delirium in a subject. The detection and monitoring of delirium may be based on the output from one or more machine learning models that process a number of features extracted from EEG signals. The output of one or more machine learning models may be used to provide or generate a subject's tendency to delirium over a period of time. If delirium is detected, any suitable therapy may be given to the subject to treat delirium (e.g., drugs, controlling the environment, addressing underlying medical conditions). The system and method may also be employed to detect and monitor other neurological conditions or brain function abnormalities, such as seizures, strokes, and sedation, and to differentiate between conditions.

[0012] In general, the present systems and methods include a machine learning model that can be trained to output a delirium positive or delirium negative assessment using one or more features of the EEG signal. The EEG features that contribute to the machine learning model may include both time domain and frequency domain characteristics.

[0013] A method for detecting and / or monitoring delirium is described herein. In one aspect, the method for detecting delirium includes acquiring data from a subject, comprising a plurality of electroencephalography (EEG) signals recorded across one or more channels or across multiple channels, and preprocessing the data by dividing the EEG signals into a plurality of time segments, each time segment corresponding to a time epoch defined by at least a start time and a duration. A plurality of features from each of the plurality of time segments may then be extracted, and one or more machine learning models may be used to generate a delirium classification for each time segment based on the extracted features. An overall delirium score for the subject may then be determined during a time window, and the overall delirium score may be based on the delirium classification generated by the one or more machine learning models and the time window including the one or more time epochs. The number of channels over which the EEG signals are recorded may range from 1 to 45, including all values ​​and subranges therein. For example, the number of channels may include 1 channel, 2 channels, 3 channels, 4 channels, 5 channels, 6 channels, 7 channels, 8 channels, 9 channels, 10 channels, 11 channels, 12 channels, 13 channels, 14 channels, 15 channels, 16 channels, 17 channels, 18 channels, 19 channels, 20 channels, 21 channels, 22 channels, 23 channels, 24 channels, 25 channels, 26 channels, 27 channels, 28 channels, 29 channels, 30 channels, 31 channels, 32 channels, 33 channels, 34 channels, 35 channels, 36 channels, 37 channels, 38 channels, 39 channels, 40 channels, 41 channels, 42 channels, 43 channels, 44 channels, or 45 channels. The multiple channels may comprise multiple electrodes, which may be coupled to or incorporated within a headband, headgear, or other device configured to place the electrodes on or around the patient's head.

[0014] A system for detecting and / or monitoring delirium is also described herein. In one aspect, the system includes a delirium detection module comprising a data module configured to receive data comprising a plurality of electroencephalography (EEG) signals recorded across one or more channels or channels during a time window from a subject, a memory storing a set of instructions, and one or more processors configured to pre-process the data received by the data module in response to the set of instructions. The pre-processing may include dividing the EEG signal into a plurality of temporal segments, each temporal segment corresponding to a time epoch defined by at least a start time and a duration, and extracting a plurality of features from each of the plurality of temporal segments. Further, one or more machine learning models may be used to generate a delirium classification for each temporal segment based on the extracted features. An overall delirium score may be determined based on the delirium classification generated by the one or more machine learning models. The number of channels in the system may range from 1 to 45, including all values ​​and subranges therein. For example, the number of channels may include 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, or 45. Multiple channels may comprise multiple electrodes, as described above, and thus the system may include a headband, headgear, or other device configured to place the electrodes on or around the patient's head.

[0015] In addition, a method for detecting brain function abnormality is disclosed herein. In one aspect, the method includes acquiring data comprising a plurality of electroencephalography (EEG) signals recorded over a plurality of channels during a time window from a subject, and pre-processing the data. The pre-processing may include dividing the EEG signal into a plurality of time-based segments, each time-based segment corresponding to a time epoch defined by at least a start time and a duration, and extracting a plurality of features from each of the plurality of time-based segments. One or more machine learning models may be used to generate a plurality of classifications for each time-based segment based on the plurality of extracted features, the plurality of classifications comprising separate classifications for each time-based segment for two or more indications selected from the group consisting of sedation, delirium, stroke, or seizure. One or more measures of brain function abnormality (BFA) may be displayed in response to the plurality of classifications.

[0016] A system for detecting BFA is further described herein. In one aspect, the system includes a BFA detection module comprising a data module configured to receive data comprising a plurality of electroencephalography (EEG) signals recorded over a plurality of channels during a time window from a subject, a memory storing a set of instructions, and one or more processors configured to pre-process the data in response to the set of instructions. The pre-processing may include dividing the EEG signal into a plurality of time-based segments, each time-based segment corresponding to a time epoch defined by at least a start time and a duration, and extracting a plurality of features from each of the plurality of time-based segments. One or more machine learning models may be used to generate a plurality of classifications for each time-based segment based on the plurality of extracted features, the plurality of classifications comprising separate classifications for two or more indications selected from the group consisting of sedation, delirium, stroke, or seizure for each time-based segment. One or more measures of BFA may be displayed in response to the plurality of classifications.

[0017] In another aspect, the present disclosure provides a neurological condition detection and monitoring system. The system may include a data module configured to acquire data comprising a plurality of electroencephalography (EEG) signals collected from a subject. The system may also include a processing module in communication with the data module. The processing module may be configured to process the data and detect and monitor one or more neurological conditions that the subject is suffering from or likely to be suffering from. The processing module may also be configured to generate an indication or assessment (i) for each neurological condition at an individual level, and optionally (ii) between two or more concomitant neurological conditions. In some cases, the one or more neurological conditions relate to at least one of sedation, delirium, stroke, or seizure.

[0018] The processing module may be configured to process the data and simultaneously detect and monitor one or more neurological conditions in real-time.

[0019] The two or more concomitant neurological conditions may include sedation and delirium. An indication or assessment generated by the processing module may indicate the degree of relationship or correlation between sedation and delirium.

[0020] The data module may include multiple electrodes configured to be placed on different regions of the subject's head, the different regions including the frontal lobe, the temporal lobe, and the occipital lobe. The data module may also include multiple channels multiplexing EEG signals from the multiple electrodes within each region and between the different regions. The data may also include non-EEG data. The non-EEG data may include one or more of the subject's blood pressure, heart rate, or motion data.

[0021] The processing module may further be configured to convert the data into one or more concomitant assessment scores based on at least the Licker Sedation-Agitation Scale (SAS), the Richmond Agitation-Sedation Scale (RASS), the Confusion Assessment Method in the Intensive Care Unit (CAM-ICU), CAM-ICU-7, the Delirium Rating Scale-Revised (DRS-R-98), the Intensive Care Delirium Screening Checklist (ICDSC), or one or more applicable scales for one or more indications.

[0022] The processing module may be further configured to generate a visual output comprising a graph displaying the probability that the subject is suffering from delirium and / or the severity of the delirium on a probability / severity scale as a function of time. The processing module may be further configured to generate one or more concomitant assessment scores indicative of the severity of the delirium. The processing module may be further configured to generate a diagnostic output based on the indication or assessment. The diagnostic output may include an aggregated wellness score or a graphical representation of the subject's brain state. The aggregated wellness score may be a combination of multiple discrete scores corresponding to multiple neurological conditions. The multiple discrete scores may be combined based on different weights allocated to the multiple neurological conditions. The graphical representation may be a combination of multiple different temporal graphs corresponding to multiple neurological conditions. The graphical representation comprises an overlay of multiple different temporal graphs.

[0023] The processing module may be configured to process the data and detect and analyze a plurality of features that are likely to be associated with a plurality of neurological conditions. The plurality of features may include a plurality of time domain features and a frequency domain feature. The plurality of features may include brain asymmetry, amplitude variation, spatial and temporal correlation, coherence, or covariates of two or more features. The plurality of features may further be ranked and classified.

[0024] The processing module is further configured to train a machine learning algorithm to use the plurality of features as input to classify different classes or severities associated with the one or more neurological conditions.

[0025] In another aspect, the present disclosure provides a method of neurological condition detection and monitoring. The method may include acquiring data comprising a plurality of electroencephalography (EEG) signals collected from a subject. The method may also include processing the data to (1) detect and monitor one or more neurological conditions to which the subject is suffering or likely to suffer, and (2) generate an indication or assessment (i) for each neurological condition at an individual level, and optionally (ii) between two or more concomitant neurological conditions. The one or more neurological conditions are selected from the group consisting of sedation, delirium, stroke, and seizures. [Brief description of the drawings]

[0026] [Figure 1] FIG. 1 illustrates a schematic diagram of a neurological condition detection and monitoring system according to an embodiment of the present disclosure.

[0027] [Diagram 2] FIG. 2 illustrates a schematic diagram of various modules for a neurological condition detection and monitoring system, according to embodiments herein.

[0028] [Diagram 3] FIG. 3 illustrates a schematic diagram of an interface of a neurological condition detection and monitoring system, according to an embodiment herein.

[0029] [Figure 4] FIG. 4 illustrates a time series plot of a neurological condition for a patient, according to embodiments herein.

[0030] [Diagram 5] FIG. 5 illustrates various meters for various neurological conditions, according to embodiments herein.

[0031] [Figure 6]FIG. 6 shows a schematic diagram of a computer system that is programmed or otherwise configured to implement the methods provided herein.

[0032] [Figure 7] FIG. 7 shows a schematic diagram of a delirium detection module according to an embodiment herein.

[0033] [Figure 8] FIG. 8 shows a schematic diagram of an alternative delirium detection module according to an embodiment herein.

[0034] [Figure 9] FIG. 9 shows a schematic diagram of another alternative delirium detection module according to embodiments herein. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0035] Detailed Description The present application relates to a system and method for detecting and monitoring delirium in a subject. Delirium monitoring may be based on the output from one or more machine learning models that process a plurality of features extracted from electroencephalography (EEG) signals. The output of one or more machine learning models may be used to provide or generate a subject's delirium tendency over a period of time. If delirium is detected, any suitable therapy may be given to the subject to treat delirium as described above (e.g., drugs, controlling the environment, addressing underlying medical conditions). The present system and method may also be employed to detect and monitor other neurological conditions or brain function abnormalities, such as seizures, strokes, and sedation, and to differentiate between conditions.

[0036] Delirium Delirium may be an acute disturbance in consciousness and cognitive function, which usually fluctuates over time, as described above. Delirium may be a common illness, with a reported incidence of over 60% during intensive care unit (ICU) stays and over 15% in geriatric or intermediate care units. Delirium may be associated with higher mortality, longer hospital stays, long-term cognitive impairment, and increased costs. Delirium is typically divided into three different subtypes based on psychomotor behavior: hypoactive, hyperactive, and mixed delirium. Hyperactive delirium may be characterized by increased motor activity, which may manifest as restlessness, agitation, aggressive behavior, wandering, and inappropriate behavior, as well as hyperarousal, hallucinations, and delusions. Hypoactive delirium may be characterized by reduced motor activity, which may manifest as lethargy, withdrawal, drowsiness, and spatial fixation. Hypoactive delirium is the most common form of delirium in older adults. Mixed delirium is characterized by subjects exhibiting aspects of both hypoactive and hyperactive delirium.

[0037] Despite its frequency and impact, delirium recognition by health care professionals is poor. Furthermore, delayed treatment of delirium in ICU patients has been found to increase mortality and morbidity. To improve early diagnosis and treatment, the American College of Critical Care Medicine and the American Psychiatric Association recommend daily monitoring of delirium in ICU patients. Various delirium clinical assessment tools have been developed. Of these, the Confusion Assessment Method for ICU (CAM-ICU) had the highest sensitivity in ICU patients. However, the sensitivity of the CAM-ICU in routine daily practice was found to be low, especially for detecting hypoactive delirium (sensitivity 31%) and delirium in postoperative patients (overall 47%). Unfortunately, the CAM-ICU has the limitation that it does not assess the severity of delirium. The Confusion Assessment Method (CAM)-ICU-7 is a surrogate assessment that provides a delirium severity scale, which is a 7-point scale (0-7) and is derived from responses to the CAM-ICU and RASS items. However, the above screening protocols are generally not well suited to the ICU behavioral patterns, which typically focus primarily on monitoring physiological changes in patients using passive methods that are not dependent on behavior-based assessments supervised by professionals.

[0038] These factors may prevent early treatment and therefore impair outcomes. Furthermore, research on delirium in the ICU may be hindered by the lack of a sensitivity tool for monitoring. Delirium can be accurately monitored using EEG and diagnosed using EEG-based biomarkers and machine learning algorithms described elsewhere herein.

[0039] In one aspect, the present disclosure provides a method for detecting delirium in a subject, comprising: acquiring data from a subject, the data comprising a plurality of electroencephalography (EEG) signals recorded across a plurality of channels; preprocessing the data by dividing the EEG signals into a plurality of temporal segments, each temporal segment corresponding to a time epoch defined by at least a start time and a duration; extracting a plurality of features from each of the plurality of temporal segments; generating a delirium classification for each temporal segment based on the plurality of extracted features using one or more machine learning models; and determining an overall delirium score for the subject for a time window, the overall delirium score being based on the delirium classification generated by the one or more machine learning models, the time window comprising one or more time epochs. In some embodiments, the delirium is hypoactive delirium.

[0040] In some embodiments, subjects may be selected based on having an increased risk for suffering from delirium. The increased delirium risk may be based on one or more of the following risk factors that may be associated with ICU admission: benzodiazepine use, blood transfusion, age, dementia, previous delirium episode, previous coma, emergency surgery, trauma, an increased Acute Physiology and Chronic Health Evaluation (APACHE) score, and an increased American Society of Anesthesiologists (ASA) Physical Status Classification System score.

[0041] In some embodiments, the delirium classification is a binary classification, delirium positive or delirium negative, a delirium probability value, or a delirium severity value. In some embodiments, the delirium classification may further classify delirium positive cases into subtypes of hypoactive, hyperactive, or diamond delirium. In some embodiments, the method further includes providing a trace of the overall delirium score over time. The method may further include determining a trend line for the trace.

[0042] In some embodiments, the data pre-processing further includes extracting a plurality of multi-channel features from different EEG signals corresponding to a given time epoch that quantify the degree of correlation between paired temporal segments, and the method further includes using a multi-channel machine learning model to generate a multi-channel delirium classification for each time epoch based on the plurality of multi-channel features, and the delirium score is further based on the multi-channel delirium classification.

[0043] In some embodiments, the time window has a duration that encompasses one time epoch, or a duration that encompasses multiple consecutive time epochs. In some embodiments, the duration of each of the time epochs is from about 1 second to about 10 minutes. The duration of each of the time epochs may be about 10 seconds, about 30 seconds, about 60 seconds, about 2 minutes, about 5 minutes, or about 10 minutes. In some embodiments, the consecutive time epochs may be non-overlapping or may overlap by 50% or less.

[0044] In some embodiments, the plurality of features comprises at least one time domain feature, at least one frequency domain feature, or at least one feature quantifying a degree of correlation between the time-based segment and a corresponding time-based segment of at least one other simultaneously collected EEG signal, which may be collected from the same hemisphere of the brain or from a different hemisphere of the brain.

[0045] In some embodiments, each channel is assigned to an independent machine learning model, and for each channel, the extracted features are applied to the machine learning model corresponding to the channel. In some embodiments, one or more of the machine learning models are random forest models. In some embodiments, the multi-channel machine learning model is a random forest model.

[0046] In one aspect, the present disclosure provides a system for detecting delirium comprising: a data module configured to receive data from a subject, the data comprising a plurality of electroencephalography (EEG) signals recorded across a plurality of channels during a time window; a memory storing a set of instructions; and one or more processors configured to preprocess the data received by the data module by dividing the EEG signals into a plurality of temporal segments, each temporal segment corresponding to a time epoch defined by at least a start time and a duration, in response to the set of instructions, extracting a plurality of features from each of the plurality of temporal segments, generate a delirium classification for each temporal segment based on the plurality of extracted features using one or more machine learning models, and determine an overall delirium score based on the delirium classification generated by the one or more machine learning models. In an embodiment, the delirium is hypoactive delirium.

[0047] In some embodiments, the delirium classification is a binary score, delirium positive or delirium negative. In some embodiments, the subject's overall delirium score is based on the percentage of time-based segments within the time window that are delirium positive, such that a higher percentage of time-based segments that are delirium positive results in a higher delirium burden or delirium severity. In some embodiments, the delirium classification comprises a delirium probability of 0 to 1. In some embodiments, the delirium classification comprises a severity value of the degree of severity of delirium.

[0048] In one embodiment, the data pre-processing further includes extracting a plurality of multi-channel features that quantify the degree of correlation between paired time-based segments from different EEG signals corresponding to a given time epoch, and generating a multi-channel delirium classification for each time epoch based on the plurality of multi-channel features using a multi-channel machine learning model, and an overall delirium score is further based on the multi-channel delirium classification.

[0049] In an embodiment, the time window has a duration that encompasses one time epoch, or a duration that encompasses multiple consecutive time epochs. In an embodiment, the time epochs have a duration of about 1 second to about 10 minutes. The time epochs may have a duration of about 10 seconds, about 30 seconds, about 60 seconds, about 2 minutes, about 5 minutes, or about 10 minutes. In an embodiment, the consecutive time epochs are non-overlapping or overlap by 50% or less.

[0050] In an embodiment, the plurality of features comprises at least one time domain feature, at least one frequency domain feature, or at least one feature quantifying a degree of correlation between the time-based segment and a corresponding time-based segment of at least one other simultaneously collected EEG signal, which may be collected from the same hemisphere of the brain or from a different hemisphere of the brain.

[0051] In one embodiment, each channel is assigned to an independent machine learning model, and for each channel, the extracted features are applied to the machine learning model corresponding to the channel. In one embodiment, one or more machine learning models are random forest models. In one embodiment, the multi-channel machine learning model is a random forest model.

[0052] In one aspect, the disclosure provides a method for detecting brain function abnormalities (BFA), comprising: acquiring data from a subject, the data comprising a plurality of electroencephalography (EEG) signals recorded across a plurality of channels for a time window; preprocessing the data by dividing the EEG signals into a plurality of time-based segments, each time-based segment corresponding to a time epoch defined by at least a start time and a duration; extracting a plurality of features from each of the plurality of time-based segments; generating a plurality of classifications for each time-based segment based on the extracted features using one or more machine learning models, the plurality of classifications comprising separate classifications for two or more indications selected from the group consisting of sedation, delirium, stroke, or seizure, for each time-based segment; and displaying one or more measures of BFA in response to the plurality of classifications.

[0053] In an embodiment, displaying the one or more measurements of BFA includes displaying separate measurements of BFA each corresponding to two or more indications. In an embodiment, displaying the one or more measurements of BFA includes displaying a combined measurement of BFA based on a plurality of classifications, comprising separate classifications for the two or more indications, for each time-based segment. In an embodiment, displaying the one or more measurements of BFA includes selecting and displaying a most likely indication from two or more indications, based on a plurality of classifications, comprising separate classifications for the two or more indications, for each time-based segment.

[0054] In one embodiment, the separate classification for each of the two or more indications comprises a binary classification: positive indication or negative indication. In one embodiment, the separate classification for each of the two or more indications comprises a probability of a subject suffering from the two or more indications within a given time epoch, ranging from 0 to 1. In one embodiment, the separate classification for each of the two or more indications comprises a severity value of the degree of severity of the two or more indications suffered by the subject.

[0055] In one embodiment, each channel is assigned to an independent machine learning model, and for each channel, the extracted features are applied to the machine learning model corresponding to the channel. In one embodiment, each of the two or more indications is assigned to an independent machine learning model, and for each of the two or more indications, the extracted features are applied to the machine learning model corresponding to the indication.

[0056] In one aspect, the present disclosure provides a system for detecting brain function abnormalities (BFA), comprising: a data module configured to receive data from a subject comprising a plurality of electroencephalography (EEG) signals recorded across a plurality of channels for a time window; a memory storing a set of instructions; and a BFA detection module comprising: a data module configured to receive data from a subject comprising a plurality of electroencephalography (EEG) signals recorded across a plurality of channels for a time window; a memory storing a set of instructions; and, in response to the set of instructions, preprocessing the data by: dividing the EEG signal into a plurality of time-based segments, each time-based segment corresponding to a time epoch defined by at least a start time and a duration; and extracting a plurality of features from each of the plurality of time-based segments, and generating a plurality of classifications for each time-based segment based on the extracted features using one or more machine learning models, the plurality of classifications comprising separate classifications for two or more indications selected from the group consisting of sedation, delirium, stroke, or seizure for each time-based segment; and one or more processors configured to display one or more measures of BFA in response to the plurality of classifications.

[0057] In an embodiment, displaying the one or more measurements of BFA includes displaying a separate measurement of BFA corresponding to each of the two or more indications. In an embodiment, displaying the one or more measurements of BFA includes displaying a combined measurement of BFA based on a plurality of classifications, the plurality of classifications comprising separate classifications for the two or more indications, for each time-based segment. In an embodiment, displaying the one or more measurements of BFA includes selecting and displaying a most likely indication from the two or more indications, the plurality of classifications, the plurality of classifications comprising separate classifications for the two or more indications, for each time-based segment.

[0058] In an embodiment, the separate classification for each of the two or more indications comprises a binary classification, either indication positive or indication negative. In an embodiment, the separate classification for each of the two or more indications comprises a probability of the subject suffering from the two or more indications within a given time epoch, between 0 and 1. In an embodiment, the separate classification comprises a severity value of the degree of severity of the two or more indications suffered by the subject. In an embodiment, each channel is assigned to an independent machine learning model, and the extracted features for each channel are applied to the machine learning model corresponding to the channel. In an embodiment, each of the two or more indications is assigned to an independent machine learning model, and the extracted features for each of the two or more indications are applied to the machine learning model corresponding to the indication.

[0059] Sedation Sedation monitoring is traditionally performed by trained health professionals, physicians, or nurses through clinical assessment using assessment scales such as the Sedation-Agitation Scale (SAS), the Richmond Agitation-Sedation Scale (RASS), or some similar variants such as the Ramsay Sedation Scale. These scales can be categorized and converted into a numerical scale, often ranging from negative numbers (highly sedated) to positive numbers (highly alert), particularly in subjective assessments. For example, the RASS ranges from -5 (unarousable) to +4 (aggressive). Using subjective clinical assessment scales to assess deeper levels of sedation can be problematic because the assessment method relies on physical patient movement or reaction, making subtle distinctions of deeper sedation levels difficult. Furthermore, the use of paralysis combined with sedation, such as during the use of a ventilator, can result in more difficult clinical assessments and can cause physician uncertainty as to whether the patient is adequately sedated.

[0060] Objective EEG-based sedation monitoring has been clinically adopted by trained anesthesiologists for use on surgical patients in the operating room (OR). In conjunction with other monitoring equipment traditionally used for sedation monitoring in the OR, anesthesiologists may be able to monitor different levels of sedation across different drug agents during surgery with greater certainty and ease than clinical assessment. There is a need for improved EEG-based sedation monitoring for widespread use outside hospital units or within hospital units such as post-anesthesia recovery rooms (PACUs) and intensive care units (ICUs). In such hospital units, sedatives may be used for a range of reasons and procedures, i.e., short-term procedures such as bronchoscopy or gastrointestinal procedures, continuous sedation for surgical patients requiring immobilization, or ventilation of patients, some of which require additional paralyzing agents. Of such situations in the ICU, continuous sedation and sedation combined with paralysis may present the greatest need for EEG-based sedation monitoring. These ICU sedation situations present challenges in accurate clinical assessment and monitoring, often resulting in over-sedation as well as under-sedation. Over-sedation in the ICU can result in reduced procedural quality, i.e., delays in treatment, longer stays, increased risk of ventilation and infection, and also impacts on neuromonitoring and triage of neurological complications such as stroke, seizures, and delirium. Under-sedation can result in significant reductions in procedural quality, primarily from severe patient discomfort from poor pain management and psychological trauma, especially during the use of paralysis. Although some use of EEG-based sedation monitoring outside the ICU exists, its use is limited due to difficulty of use, lack of easy-to-understand readouts, and the resulting need for highly trained professionals to operate the EEG.

[0061] Sedation and delirium The independent clinical assessment and treatment challenges for both sedation and delirium can be further complicated by the relationship that exists between patient sedation and delirium; that is, excessive sedation and / or the use of certain sedatives can increase a patient's risk of delirium. Nonetheless, sedation may sometimes be the preferred treatment method for some manifestations of delirium (i.e., hyperactive delirium), which may prolong the effects of other delirium subtypes. Furthermore, sedation, especially excessive sedation, may mask symptoms of delirium present at the time of clinical awareness, limiting clinical feedback of whether treatment is effective. As a result, there is a need to effectively monitor and detect both a patient's sedation and delirium levels. Detection and Surveillance Systems I. Signal Acquisition and Preprocessing II. Signal analysis III. Neurological Condition Detection and Output IV. Delirium Detection V. After neurological condition detection VI. Computer Systems

[0062] I. Signal Acquisition and Preprocessing For ease of explanation, the figures and corresponding description below are described below with reference to the analysis of signals representative of brain activity (e.g., electroencephalography (EEG) signals) and / or cardiac activity (e.g., electrocardiogram (ECG) signals) of a living subject. However, those skilled in the art will recognize that signals representative of other bodily functions (e.g., electromyography (EMG) signals, or electronystagmography (ENG) signals, pulse oximetry signals, capnography signals, and / or photoplethysmography signals) may be substituted or used in addition to (e.g., in conjunction with) one or more signals representative of brain activity and / or cardiac activity. In some variations, the signal is an EEG signal that is analyzed to detect delirium in a patient.

[0063] A system for measuring bioelectrical signals may generally include one or more electrodes electrically coupled to a controller and / or output device via corresponding conductive wires. In other variations, the electrodes may be wirelessly coupled to the controller and / or output device. The electrodes may be contained within an electrode carrier system that is secured around the patient's head. The electrode carrier system may be configured as a headband or incorporated into any number of other platforms or positioning mechanisms for maintaining the electrodes against the patient's body. The individual electrode assemblies may be spaced apart from one another so that when the headband is positioned on the patient's head, the electrode assemblies may be optimally aligned to receive EEG signals. In some variations, the electrode carrier system may be used to detect delirium in a patient.

[0064] In some variations, EEG signals from 10 electrodes may be combined. The electrode locations may be, for example, Fp1, Fp2, F7, F8, T3, T4, T5, T6, O1, and O2. These electrodes may form 8 channels (Fp1-F7, F7-T3, T3-T5, T5-O1, Fp2-F8, F8-T4, T4-T6, and T6-O2, or any combination thereof). In other variations, EEG signals from 16 electrodes may be combined. The electrode locations may be, for example, Fp1, Fp2, F3, F4, F7, F8, C3, C4, P3, P4, T3, T4, T5, T6, O1, and O2. The use of 16 electrodes may be employed to generate long-field EEG channels useful for detecting delirium. For example, EEG electrodes Fp1-F7, F7-T3, and T3-T5 may be used to generate long-field channel Fp1-T5. This extended EEG montage may be used for subsequent processing and prediction for delirium detection. Other long-field channels that may be generated include, but are not limited to, Fp1-O1, Fp1-T5, F7-O1, Fp1-T3, F7-T5, T3-O1, Fp2-O2, Fp2-T6, F8-O2, Fp2-T4, F8-T6, T4-O2, Fp1-F8, Fp1-T4, Fp1-T6, Fp1-O2, F7-Fp2, F7-F8, F7-T4, F7-T6, F7-O2, T3-Fp2, T3-F8, T3-T4, T3-T6, T3-O2, T5-Fp2, Including T5-F8, T5-T4, T5-T6, T5-O2, O1-Fp2, O1-F8, O1-T4, O1-T6, Fp1-Fp2, and O1-O2.

[0065] The number of channels from which EEG signals are acquired and recorded may range from 1 to 45, including all values ​​and subranges therein. For example, the multiple channels may include 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, or 20, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, or 45 channels. In one variation, an 8-channel EEG may be used to detect delirium. In another variation, a 16-channel EEG may be used to detect delirium.

[0066] The electrode may be part of an electrode assembly and electrode carrier system, as described above. The electrode carrier system may generally include an electrode body that is at least partially conductive, one or more members (e.g., one or more tubular members) extending from the electrode body, each defining a lumen therethrough and a distal opening, a reservoir having a compressible structure and containing a conductive fluid or gel in fluid communication with the one or more members, and a substrate supporting the electrode body and the reservoir.

[0067] In some variations, the electrode carrier system may generally comprise an electrode body having one or more tubular members extending therefrom, each defining a lumen therethrough and a distal opening, a reservoir having a compressible structure defining an internal volume and in fluid communication with the one or more tubular members, and a controller and / or output device in electrical communication with the electrode body, the controller and / or output device configured to receive electrical signals from the electrode assembly and record and / or output a corresponding response.

[0068] The electrode carrier system may generally comprise a substrate that is secured around the patient's head. The substrate may be configured as a headband, although the carrier system may be incorporated into any number of other platforms or positioning mechanisms for maintaining the electrodes against the patient's body. The individual electrodes are spaced apart from one another such that when the headband is positioned on the patient's head, the electrodes are optimally aligned on the head for receiving EEG signals. The carrier system may electrically couple each of the electrodes via corresponding conductive wires extending from the substrate, for example, to a controller and / or output device. However, in other variations, the electrodes may be wirelessly coupled to the controller and / or output device.

[0069] The controller and / or output device may generally comprise any number of devices for receiving electrical signals, such as electrophysiological monitoring devices, and may be used in combination with any number of brain imaging devices, e.g., fMRI, PET, NIRS, etc. In one particular variation, the electrode embodiments described herein may be used in combination with devices such as those configured to receive electrical signals from the electrodes and process them.

[0070] In one variation, the electrode carrier system may comprise each of the electrodes enclosed within a reservoir that is pre-filled with a conductive gel or fluid. Each electrode may be configured in a flattened or atraumatic configuration contained within an individual reservoir, and each reservoir may be formed from any number of easily crushable flexible materials, such as silicone, polyurethane, rubber, etc. The electrodes may be coupled via a conductive wire passing through a lumen defined through the substrate, separated from the electrodes by a substrate. Each reservoir may also respectively define one or more openings through which the conductive gel or fluid may be drained.

[0071] Once the platform is seated over the patient's head, the user may press on each of the reservoirs such that the conductive fluid or gel flows through the openings and onto the patient's skin. The conductive fluid or gel that is expelled through the openings may maintain fluid communication between the skin surface and the individual electrodes such that detected electrical signals may be transmitted from the skin and the electrodes. Furthermore, due to the flexibility of the reservoirs, once the conductive fluid or gel is expelled into contact with the skin surface, the substrate may be placed flat against the skin surface such that the patient may comfortably place their head on the surface while still maintaining electrical contact with the electrodes.

[0072] Another electrode variation may consist of one or more loops of conductive wire or ribbon that can be easily bent or flexed relative to the skin surface. The electrode carrier system may include a pressure relief reservoir around each of the electrodes to contain a conductive fluid or gel, as described above, so that the conductive fluid or gel may drain around and into the one or more loops to ensure a conductive pathway.

[0073] In further variations of the electrode assembly, one or more tubular members may extend laterally from the substrate. The tubular members may be arranged in a circular pattern, each for each electrode, and they may also define a lumen therethrough, with an opening defined at each distal end. Each tubular member may be fabricated from a conductive metal, which may retain its tubular shape in use, or may be thin and flexible enough to bend or yield when placed against the patient's skin surface. Alternatively, the tubular members may be fabricated from a flexible material, which is coated or layered with a conductive material so that the members retain their flexibility. In either case, the conductive fluid or gel may be contained within the tubular member, or they may be reserved in a pressure relief reservoir surrounding or adjacent to each electrode, as described above. Due to the tubular shape of the electrodes, they may easily pass through the patient's hair, if present, and contact against the skin surface while maintaining electrical contact.

[0074] Yet another variation of the electrode embodiment may also utilize a pressure-relief reservoir that is filled with a conductive fluid or gel. The reservoir may be formed from a flexible material, such as silicone, polyurethane, rubber, etc., that extends from a substrate, forming a curved or arc-like structure, with one or more openings defined across the reservoir. These openings may remain closed until a force is applied to the reservoir and / or substrate, which may force the fluid or gel contained therein to escape through the openings and contact the exterior surface of the reservoir and the underlying skin surface. The exterior surface of the reservoir may have a layer of conductive material in electrical contact with the conductive wires, such that electrical contact may be achieved once the fluid or gel is expelled from within the reservoir and out onto the conductive material on the exterior surface of the reservoir and onto the skin surface.

[0075] The electrode carrier system may include an electrode body that, in some cases, defines one or more tubular members extending from the body such that the members project laterally away from the substrate. The electrode body may be made of a conductive material such as metal, and may be rigid. However, in other variations, the body may be fabricated from a conductive material that is also flexible, such as conductive silicone, and / or from a flexible material, such as silicone, polyurethane, rubber, etc., that may be coated or layered with a conductive material so that the underlying tubular members retain their flexibility. In either case, the body may be affixed to the substrate such that one or more openings defined along the body and extending through the member are in fluid communication with a reservoir having a compressible housing. The reservoir may also be affixed to the substrate and contain a volume of conductive fluid or gel locally in the electrode body.

[0076] The tubular members may be arranged in a variety of patterns, for example, circular, uniform, or arbitrary. When the substrate is adhered to a patient, the reservoirs may be pressed or urged such that the fluid or gel contained therein is expelled through each of the tubular members and through the corresponding distal opening into contact with the underlying skin surface. The elongated nature of the members may allow them to easily pass through the patient's hair, if present, and make direct contact with the skin surface.

[0077] In another variation, the electrode carrier system having a tubular body may define one or more openings across its surface. The tubular body may have one or more tubular members that extend in a spiral or helical pattern away from the substrate. The tubular member may define a lumen therethrough that extends from the tubular body and a distal opening at its tip. The substrate may further define a reservoir that contains a volume of conductive fluid or gel such that the body is in fluid communication with the reservoir. In some variations, the distal tip of the member may present a roughened surface for contacting the skin. Optionally, the roughened tip may be rotated or otherwise translated or traversed by the user across the skin surface to at least partially ablate the skin surface, facilitating electrical contact.

[0078] For example, the distal skin-contacting surface of the electrode assembly may be modified to prepare the skin surface and improve electrical conductivity (i.e., lower electrical resistance) between the conductive portion of the electrode assembly and the skin when the conductive portion thereof is in physical contact with the skin. For example, the tissue-contacting surface of the electrode assembly may be modified to have an abrasive surface, e.g., by coating with abrasive particles, may be formed or shaped to have protruding rigidity features, e.g., bumps, ridges, or the like, and / or may be coated with a material that lowers the electrode connection impedance. Such sweeping and / or chemical coating of the tissue-contacting surface of the electrode assembly over the target tissue location may scrape, dissolve, and / or otherwise disrupt dead tissue and break down scalp oils. In specific examples, at least a portion of the distal tissue-contacting surface of the electrode assembly, e.g., the distal surface of at least some of the tubular members, comprises such surface features, surface coatings, surface treatments, or combinations thereof to improve the quality of the electrode connection.

[0079] In another specific aspect of the invention, an electrode assembly comprises an electrode body and one or more tubular members extending from the electrode body, typically from a bottom surface of the electrode body. Each tubular member has a distal tip, and at least some of the tubular members have a lumen with a distal opening at the distal tip. A reservoir containing a conductive fluid or gel is optionally disposed within the electrode body, and the electrode body is configured for dispensing the conductive fluid or gel from the reservoir, through the lumen, and out of the distal opening of the tubular member. Alternatively, in some embodiments, the conductive fluid or gel may be dispensed onto or through the lumen of the tubular member using a syringe or other separate delivery device.

[0080] In specific embodiments, the electrode assembly typically comprises at least two tubular members, and may comprise three tubular members, four tubular members, or even more. The tubular members will typically depend vertically downward from the bottom surface of the electrode body, and will be specifically configured such that they can penetrate the patient's hair such that the distal tips of the tubular members will be able to engage and provide a secure electrical contact with the patient's scalp. The tissue-engaging area of ​​the tubular members on the bottom surface of the electrode body will typically be 50% or less of the area of ​​the bottom surface, frequently 30% or less of the area of ​​the electrode body, and typically at least 5% of the area of ​​the bottom surface. Thus, the tissue-engaging area of ​​the tubular members on the bottom surface of the electrode body will typically be in the range of 5% to 50% of the area of ​​the bottom surface, typically in the range of 5% to 30% of the area of ​​the bottom surface.

[0081] In most cases, the tubular members will extend from the generally planar bottom of the electrode body at a vertical angle. However, in other cases, the tubular members may extend at an angle anywhere between 30° and 150° relative to the plane, typically between 60° and 120° relative to the plane. However, in other cases, the tubular members may have other configurations, such as being configured in a helical shape, such that they can pierce the hair and reach the patient's scalp by rotating the electrode assembly about a vertical axis.

[0082] In other specific embodiments of the invention, the distal tips of at least some of the tubular members will have a skin preparation, e.g., a tissue-roughened surface. For example, the tissue-roughened surface may comprise an abrasive material, such as grit or other abrasive particles, formed over at least a portion of the distal tips of the tubular members. In other cases, the surface roughening may comprise surface features, such as ridges, bumps, grooves, and the like, formed over at least a portion of the distal tips that contact the patient's skin.

[0083] The electrode body, and in particular the tubular member connected to the electrode body, may be formed at least in part from a conductive material, such as a metal, a conductive coating, an embedded wire, or a conductive polymer. In such cases, the electrode body and / or the tubular member will provide at least a portion of the electrical pathway required to conduct a biological current from the tip of the tubular member to an electrical terminal or other conductive connector on the electrode body, as described below. However, in other cases, the electrode body and / or the tubular member may be formed primarily, or even completely, from an electrically non-conductive material. In such cases, a conductive fluid or gel will provide most or all of the conductive pathway required to deliver a biological current from the distal tip of the tubular member to the electrical terminal after such conductive fluid or gel is dispersed throughout the electrode body and the tubular member.

[0084] The members may comprise a variety of geometric shapes. In some cases, the tubular members may be generally cylindrical with a lumen extending therethrough. In other cases, however, the tubular members may be formed as "prongs" at their distal tips, having a relatively extensive tissue-contacting surface area along a curved "axis." In many cases, the tissue-contacting surface area of ​​the prongs will be generally crescent-shaped, such that as they are rotated relative to the patient's tissue, they will follow a generally circular path.

[0085] The prongs and other members (e.g., tubular members) may have ports in their tissue-contacting surfaces for delivering conductive fluids or gels to the patient's skin. In some cases, the ports may be formed in a generally flat bottom surface of the tubular member or prong. In other cases, the ports may be connected to channels or other dispersive features on the tissue-contacting surface of the prong or other tubular member. In still further specific embodiments, the ports for delivering conductive fluids or gels may be located in a recessed surface of the prong, which may be adjacent to the tissue-contacting surface below the prong or other tubular member.

[0086] An electrode assembly will typically include one or more members (e.g., prongs, tubular members) as just discussed, but in some alternative embodiments, the electrode body may have a generally flat bottom, free of tubular and other protruding members. The flat bottom may be configured to engage the skin and have an opening for releasing the conductive fluid or gel in any of the ways described elsewhere herein for delivering the conductive fluid or gel. The tissue contacting surface of such a flat bottom may be modified, e.g., roughened or textured, and electrically conductive with the target tissue surface in any of the ways discussed herein.

[0087] In use, the electrodes may be placed on the patient's scalp by placing a headband or other headgear around the patient's scalp. For example, as described above, a headband carrying the electrode assemblies and the distal tips of one or more tubular members extending from at least some of the electrode assemblies may be engaged against scalp tissue. A conductive fluid or gel may then be extruded from a reservoir disposed within at least some of the electrode assemblies such that the fluid or gel passes through the tubular members and forms a conductive pathway to the patient's scalp tissue. The electrode assemblies may then be connected to a controller and / or output device configured to receive a low-power biological current from the electrode assemblies. In some variations, at least some of the electrode assemblies have a roughened surface that may be rotated to abrade the scalp tissue adjacent the distal tips of the one or more tubular members to reduce contact resistance between the electrode assemblies and the scalp tissue.

[0088] In some embodiments, the signal corresponding to the brain electrical activity corresponds to an electrical signal obtained from a human brain and obtained from a single neuron or from multiple neurons. In some embodiments, the sensor includes one or more sensors (e.g., extracranial sensors) that are externally attached (e.g., taped, attached, glued) to the human scalp. For example, the extracranial sensor may include an electrode (e.g., electroencephalography (EEG) electrode) or multiple electrodes (e.g., electroencephalography (EEG) electrodes) that are externally attached to the scalp (e.g., glued to the skin and using a conductive gel to form an electrical contact), or more commonly, positioned at a discrete location outside the scalp. Alternatively, dry electrodes can also be used in some implementations (e.g., conductive sensors that are mechanically placed against the body of a living subject rather than embedded within the body of the living subject or contacted through a conductive gel). An example of a dry electrode is a headband with one or more metal sensors (e.g., electrodes) that are worn by the living subject during use. Signals obtained from extracranial sensors may sometimes be referred to as EEG signals or time-domain EEG signals. In some cases, the sensors may be accelerometers or inertial measurement units (IMUs) that may measure mechanical movement of the subject and / or device (e.g., produce one or more electrical signals that correspond to mechanical movement of the subject and / or device). The system may be configured to utilize one or more sensors to aid in determining potential neurological conditions as described elsewhere herein.

[0089] Neurological Condition Detection and Monitoring System - Data Module In one aspect, the present disclosure provides a neurological condition detection and monitoring system 100. As shown in FIG. 1, the system 100 may include a data module 110 configured to acquire data. The data acquired by the data module 110 may include a plurality of electroencephalography (EEG) signals collected from a subject. The data may also include non-EEG data. The non-EEG data may include blood pressure, heart rate, and / or motion data of the subject. The non-EEG data may be as described anywhere herein.

[0090] In another aspect, the present disclosure provides a neurological condition detection and monitoring method. The neurological condition detection and monitoring method may include acquiring data. The data may include a plurality of electroencephalography (EEG) signals collected from a subject. The method may include processing the data to (1) detect and monitor one or more neurological conditions that the subject is suffering from or likely to suffer from, and (2) generate an indication or assessment (i) for each neurological condition at an individual level, and optionally (ii) between two or more concomitant neurological conditions, where the one or more neurological conditions are selected from the group consisting of sedation, delirium, stroke, and seizures.

[0091] The data module 110 may include multiple electrodes configured to be placed on different regions of the subject's head. The different regions may include the frontal lobe, the temporal lobe, and / or the occipital lobe. The data module may also include multiple channels that multiplex the EEG signals from the multiple electrodes within each region and between the different regions. The electrodes may be used on the patient's frontal lobe. The electrode locations may be, for example, Fp1, Fp2, F7, F8, T3, T4, T5, T6, O1, O2, and channels (Fp1-F7, F7-T3, T3-T5, T5-O1, Fp2-F8, F8-T4, T4-T6, and T6-O2, or any combination thereof. In some cases, having electrodes at multiple locations may provide broader coverage, more channels, higher tolerance to noise or artifacts from a particular channel, and being able to monitor the effects of different agents affecting different parts of the brain. In some cases, having electrodes at multiple locations may allow for more accurate or precise determination of a neurological condition. The number of channels provided by an electrode may range, for example, from 1 to 45, as described herein above. When delirium is to be detected, 8 or 16 channels may be used. For example, when used for delirium detection, electrodes Fp1-F7, F7-T3, and T3-T5 may be used to generate long field channels Fp1-T5. In other variations for delirium detection, long field channels that may be generated include, but are not limited to, Fp1-O1, Fp1-T5, F7-O1, Fp1-T3, F7-T5, T3-O1, Fp2-O2, Fp 2-T6, F8-O2, Fp2-T4, F8-T6, T4-O2, Fp1-F8, Fp1-T4, Fp1-T6, Fp1-O2, F7-Fp2, F7-F8, F7-T4, F7-T6, F7-O2, T3-Fp2, Including T3-F8, T3-T4, T3-T6, T3-O2, T5-Fp2, T5-F8, T5-T4, T5-T6, T5-O2, O1-Fp2, O1-F8, O1-T4, O1-T6, Fp1-Fp2, O1-O2.

[0092] In some embodiments, the data module 110 may have one or more analog front ends configured to receive sensor EEG signals from the sensors. The EEG signals may be pre-processed as described elsewhere herein. In some embodiments, separate (e.g., independent) analog front ends may be provided to interface with each of the sets of sensors. In some embodiments, one or more analog front ends may be provided to interface with the set of EEG sensors.

[0093] Neurological Condition Detection and Monitoring System - Processing Module The system may include a processing module 120 in communication with the data module 110. The processing module 120 may be configured to process the data (e.g., EEG signals) and detect and monitor one or more neurological conditions that the subject is suffering from or likely to be suffering from. The processing module 120 may generate an indication or assessment (i) for each neurological condition at an individual level, and optionally (ii) between two or more concomitant neurological conditions. The indication or assessment may be presented to the user via a notification output module 290.

[0094] 2, the processing module 120 may be configured to process data from the data module 110 in real time and simultaneously detect and monitor one or more neurological conditions. The one or more neurological conditions may be associated with at least one of sedation, delirium, stroke, and seizure, for example. The two or more concomitant neurological conditions may include sedation and / or delirium. An indication or assessment generated by the processing module may indicate the degree of relationship or correlation between sedation and delirium. The processing module may be configured to convert the data into one or more contingent assessment scores based on at least the Licker Sedation-Agitation Scale (SAS), Richmond Agitation-Sedation Scale (RASS), Bispectral Index Monitor (BIS), and / or one or more applicable scales for the Confusion Assessment Method in Intensive Care Units (CAM-ICU), CAM-ICU-7, Delirium Rating Scale-Revised (DRS-R-98), Intensive Care Delirium Screening Checklist (ICDSC), and / or one or more indications. The processing module may be configured to convert the data into one or more contingent assessment scores based on at least the Ramsay Sedation Scale (RSS).

[0095] The processing module may be configured to convert the data into one or more contingency assessment scores based on at least the Licker Sedation-Arousal Scale (SAS). The SAS scale may range from 1 to 7. A 1 on the SAS scale may indicate an unarousable state, where the subject has little or no response to noxious stimuli. The subject is unable to communicate or follow commands. A 2 on the SAS scale may indicate a highly sedated state, where the subject may be aroused to physical stimuli but does not communicate or follow commands. The subject may be mobile on their own. A 3 on the SAS scale may indicate a sedated state, where the subject is difficult to arouse but is aroused to verbal stimuli or gentle rocking, follows simple commands, but falls asleep again. A 4 on the SAS scale may indicate a calm and cooperative state, where the subject is calm, easily arousable, and follows commands. A 5 on the SAS scale may indicate an aroused state, where the subject feels anxious or physically aroused but calms down to verbal commands. A 6 on the SAS scale may indicate a very agitated state, with the subject needing to be restrained, frequent verbal reminders of limitations, biting the ETT, etc. A 7 on the SAS scale may indicate a dangerously agitated state, with the subject pulling on the ET tube, trying to remove the catheter, climbing over the bed rails, attacking staff, swaying from side to side, etc. In some cases, the subject is scored at the most severe level of agitation displayed.

[0096] The processing module may be configured to convert the data into one or more contingent assessment scores based on at least the Richmond Arousal Sedation Scale (RASS). The RASS scale may range from -5 to +4. A score of negative 5 on the RASS scale may indicate non-arousal, with no response to audio or physical stimuli. A score of negative 4 on the RASS scale may indicate deep sedation, with no response from the subject to audio but moving or opening eyes to physical stimuli. A score of negative 3 on the RASS scale may indicate moderate sedation, with the subject moving or opening eyes to audio (but not making eye contact). A score of negative 2 on the RASS scale may indicate light sedation, with the subject briefly awakening and making eye contact to audio (<10 seconds). A negative score of 1 on the RASS scale may indicate somnolence, where the subject is not fully alert but is persistently aroused (eyes open / eye contact) to sounds (>10 seconds). A zero score on the RASS scale may indicate alert and calm. A positive score of 1 on the RASS scale may indicate restlessness, where the subject feels anxious but is not aggressive or active in movement. A positive score of 2 on the RASS scale may indicate agitation, where the subject has frequent purposeless movements and resists the ventilator. A positive score of 3 on the RASS scale may indicate a very agitated state, where the subject pulls or pulls out tubes or catheters, i.e., is aggressive. A positive score of 4 on the RASS scale may indicate an aggressive state, where the subject is overly aggressive, violent, and poses an immediate danger to staff.

[0097] The processing module may be configured to convert the data into one or more contingent assessment scores based at least on a bispectral index monitor (BIS). The BIS scale may be 0 to 100. A value of 0 may represent an absence of brain activity. A value of 100 may represent an awake state. A BIS value of 40 to 60 may represent proper general anesthesia for surgery, and a value below 40 may represent a deep hypnotic state.

[0098] The processing module may be configured to convert the data into one or more contingency assessment scores based on at least the Ramsay Sedation Scale (RSS). The RSS scale may range from 1 to 6. A 1 on the RSS scale may indicate that the patient is anxious and agitated or agitated or both. A 2 on the RSS scale may indicate that the patient is cooperative, alert, and calm. A 3 on the RSS scale may indicate that the patient responds only to commands. A 4 on the RSS scale may indicate that the patient exhibits an acute response to a light forehead tap or a loud auditory stimulus. A 5 on the RSS scale may indicate that the patient exhibits an insensitive response to a light forehead tap or a loud auditory stimulus. A 6 on the RSS scale may indicate that the patient exhibits no response at all.

[0099] The processing module may be configured to convert the data into one or more contingency assessment scores based on at least the Confusion Assessment Method in Intensive Care Units (CAM-ICU) or CAM-ICU-7, the Delirium Rating Scale-Revised (DRS-R-98), or the Intensive Care Delirium Screening Checklist (ICDSC). The CAM-ICU flowsheet may depend on the acute changes or fluctuating course of the patient's mental status, the subject's attention deficit, the subject's altered level of consciousness, and the subject's disorganized thinking. The CAM-ICU-7 provides a 0-7 scale for the severity of delirium.

[0100] Preprocessing of EEG signals In some embodiments, the method may include preprocessing the multiple signals by segmenting the multiple signals per channel into multiple temporal data segments. As shown in FIG. 2, the processing module 120 may include a preprocessing module 210. In some embodiments, the method may include preprocessing the multiple EEG signals by segmenting the multiple EEG signals per channel into multiple temporal data segments. FIG. 2 shows an illustration of the processing module 120. The processing module takes the EEG signals from the multiple channels from the data module 110. The processing module may preprocess the EEG signals from the multiple channels using a preprocessing module 210 configured to preprocess the EEG signals. As shown in FIG. 2, the preprocessing module may include a signal filtering module 215, a signal segmentation module 220, and a signal conditioning module 225.

[0101] In some embodiments, the filtering module 215 may be configured to filter the EEG signals from the incoming set of channels from the EEG device module, as described elsewhere herein. In some cases, the pre-processing may include, for example, segmenting the EEG signals, filtering the EEG signals based on frequency, conditioning the EEG signals, or as described elsewhere herein.

[0102] In FIG. 2, the signal segmentation module 220 can be configured to segment the EEG signal. Each temporal data segment (which may be referred to herein as a "temporal segment") of the EEG signal may be associated with a given time epoch. Each time epoch may be defined by a start time and a duration. Multiple EEG signals from different EEG electrodes may be segmented to have temporal segments corresponding to the same time epoch so that they may be analyzed, as an example, to subsequently detect temporal correlations in their respective waveforms. For each corresponding time epoch, a cluster of neurological condition positive classifications based on an analysis of features extracted from the corresponding temporal segments, as described herein below, may indicate one or more potential neurological conditions. The neurological conditions may be for delirium, seizures, sedation, stroke, and / or any combination thereof.

[0103] In some embodiments, the plurality of EEG signals may be segmented into 1 to 100,000 data segments. In some cases, the number of EEG data segments may depend on the duration of the EEG recording. In some cases, the number of EEG data segments may be fixed regardless of the duration of the EEG recording.

[0104] In some embodiments, each temporal data segment may have a duration of about 1 second to 1 hour. In some cases, each temporal data segment may have a duration of about 1 second to 30 seconds. In some cases, each temporal data segment may have a duration of about 1 second to 10 seconds. In some cases, the duration of each temporal data segment may be fixed throughout the EEG recording. In some cases, the duration of each temporal data segment may be variable or adaptive during the EEG recording.

[0105] In some embodiments, pre-processing the plurality of EEG signals may include applying one or more filtering steps to the plurality of EEG signals via the plurality of channels. Pre-processing the plurality of EEG signals may include using at least one filter, two filters, three filters, four filters, five filters, six filters, seven filters, eight filters, nine filters, ten filters, fifteen filters, or more filters. Pre-processing the plurality of EEG signals may include using up to about fifteen filters, ten filters, nine filters, eight filters, seven filters, six filters, five filters, four filters, three filters, two filters, or less. Pre-processing the plurality of EEG signals may include using any number of 1-15 filters, 1-10 filters, 1-5 filters, or 1-3 filters.

[0106] In some embodiments, one or more filtering steps may be applied before, during, and / or after segmentation of the EEG signals. One or more of the filtering steps may include, for example, a digital filter, a similar filter, or a combination thereof. One or more of the filtering steps may include, for example, a band pass filter, a low pass filter, a high pass filter, a band reject filter, an all pass filter, a Kalman filter, an adaptive filter, or a notch filter, etc. In some cases, the low frequency cutoff of the filter may be 0.1 Hz to 5 Hz. In some cases, the high frequency cutoff of the filter may be 5 Hz to 200 Hz. In some cases, the notch filter frequency may match the local power line frequency. In some cases, the notch filter frequency may be 50 Hz or 60 Hz to match the local power line frequency.

[0107] For delirium detection, a 16-channel EEG may be generated from raw waveforms recorded at a sampling rate of 250 Hz. The signal may be band-pass filtered from 0.5 to 40 Hz using a 5th order Butterworth filter. In some variations, subsequent high-pass filtering may be used for better removal of DC components for some of the feature calculations, as further described below. The filtered EEG data may then be segmented into window sizes (durations) ranging from about 1 second to about 10 minutes, including all values ​​and subranges therein. In some instances, the window size may be greater than 10 minutes. The window size may be about 1 second, about 2 seconds, about 3 seconds, about 4 seconds, about 5 seconds, about 10 seconds, about 15 seconds, about 20 seconds, about 25 seconds, about 30 seconds, about 35 seconds, about 40 seconds, about 45 seconds, about 50 seconds, about 55 seconds, about 60 seconds, about 2 minutes, about 3 minutes, about 4 minutes, about 5 minutes, about 6 minutes, about 7 minutes, about 8 minutes, about 9 minutes, or about 10 minutes, with 0% to 95% overlap (between successive windows). For example, in some variations, the filtered EEG data may be segmented into 15 second windows, with 33% (5 seconds) overlap between successive windows. In other variations, the filtered EEG may be segmented into 10 second windows with no overlap, or 4 second windows with 75% (3 seconds) overlap, or 60 second windows with 50% (30 seconds) overlap.

[0108] EEG signal conditioning 2 illustrates a signal conditioning module 225 configured to condition an EEG signal. In some embodiments, the method may condition any EEG signal. Conditioning an EEG signal may include, for example, increasing and / or decreasing the amplitude of the EEG signal, adding or decreasing the noise level of the EEG signal, increasing and / or decreasing the time epoch of the EEG signal, increasing and / or decreasing the intensity of the EEG signal, increasing and / or decreasing the signal frequency of the EEG signal, increasing and / or decreasing the voltage of the EEG signal, changing the morphological structure of the EEG signal (e.g., the shape of the EEG signal), increasing and / or decreasing the periodicity of the EEG signal, increasing or decreasing the synchrony of the EEG waves, spectral subtraction, normalization, etc.

[0109] In some cases, the EEG signal may be reduced. In some cases, the EEG signal may be downsampled to a lower sampling frequency. For example, EEG data recorded at a sampling frequency of 500 Hz may be downsampled by a factor of two to 250 Hz.

[0110] In some cases, the EEG signal may undergo bit-width reduction. In some cases, the level of resolution at which the EEG signal is recorded may not be required by the method to achieve accurate neurological condition detection. In some cases, bit-width reduction may reduce the EEG signal to a lower number of bits per sample, for example, from 32 bits per sample to 12 bits per sample, through standard quantization of the EEG signal. In some cases, bit-width reduction may be advantageous because it may be useful to reduce power consumption due to reduced processing load when the method is implemented in a portable system.

[0111] In some cases, spectral subtraction may be used to reduce the amount of additive noise in the EEG signal. In some cases, the noise may be caused by the external surroundings. In some cases, the noise may be caused by the measurement equipment. In some cases, the noise may be caused by the user. In some cases, the average frequency spectrum of the non-stroke EEG signal may be calculated to provide a base level estimate of the noise frequency spectrum over a period of time. In some cases, the EEG signal may be transformed into the frequency domain as the EEG signal is recorded. In some cases, the average noise spectrum may then be subtracted from the EEG frequency spectrum. In some cases, the resulting spectrum and phase information from the original noise signal may be combined. In some cases, the resulting spectrum may be transformed back to the time domain to produce a denoised signal.

[0112] In some embodiments, the EEG signals may be standardized by eliminating the effect of the montage used in collecting the EEG signals. In some cases, independent component analysis (ICA) or principal component analysis (PCA) methods may be used to provide montage rejection. In some cases, ICA or PCA methods may separate the EEG signals into a set of sources independent of the montage used to record them. In some cases, using standardized EEG data may remove errors introduced by the varying practices of clinicians.

[0113] In some cases, a non-negative matrix factorization (NMF) method may be applied to each channel as a form of artifact removal. In some cases, the spectrum of the signal may be decomposed into extracted bases to obtain weights. In some cases, the spectrum may be reconstructed using the bases and corresponding weights of the artifacts removed from the initial EEG signal. In some cases, an independent component analysis (ICA) or principal component analysis (PCA) method may be used for artifact reduction or removal.

[0114] II. Signal analysis The processing module 120 may be configured to process the data and detect and analyze multiple features likely to be associated with multiple neurological conditions. The processing module may extract relevant new features from the data (e.g., EEG) that may be used to estimate, by way of example, the Licker Sedation-Agitation Scale (SAS), Richmond Agitation-Sedation Scale (RASS), Bispectral Index Monitor (BIS), Confusion Assessment Method in Intensive Care Units (CAM-ICU), CAM-ICU-7, Delirium Rating Scale-Revised (DRS-R-98), Intensive Care Delirium Screening Checklist (ICDSC), or other scales. The processing module 120 may be configured to train a machine learning algorithm using the multiple features as input to classify different classes or severities associated with multiple neurological conditions.

[0115] Feature Extraction 2, the processing module may include a signal analysis module 240. The signal analysis module 240 may include a feature extraction module 245 and a machine learning classification module 250. The feature extraction module 245 may be configured to obtain pre-processed measured data (e.g., EEG signals or temporal segments of EEG signals for a given time epoch) from the pre-processing module 210 and construct derived values ​​(e.g., features).

[0116] The features extracted by the feature extraction module 245 may include time domain features and frequency domain features. The features may include time domain features and frequency domain features of the data provided from the data module 110 and / or the pre-processing module 210. The features may include brain asymmetry, amplitude variability, spatial and temporal correlation, coherence, or covariates of two or more features. The features may be ranked and / or categorized.

[0117] In some embodiments, feature extraction may start from an initial set of measured data (e.g., EEG signals for a given time epoch or temporal segments of EEG signals, etc.) and build derived values ​​(e.g., features) that are intended to be informative and non-redundant. In some cases, the feature extraction module may include extracting multiple features from each temporal data segment for each channel individually. In some cases, the feature extraction module may include extracting multiple features from each temporal data segment for all channels together. In some cases, the feature extraction module may include extracting multiple features from each temporal data segment of one or more groups, each group consisting of one or more channels.

[0118] 2, the extracted features can be relayed to a machine learning classification module 250, which can be configured to analyze and classify the extracted features as described elsewhere herein. In some cases, feature extraction can facilitate subsequent learning and generalization steps of a machine learning algorithm. In some cases, feature extraction can lead to better human interpretation. In some cases, feature extraction can be associated with dimensionality reduction.

[0119] In some cases, when the input data to a machine learning algorithm (e.g., EEG signals) is suspected to be too large to process and redundant (e.g., repetition of the same measurements or characteristics in both Hz and seconds), the data can be transformed into a reduced set of features.

[0120] In some cases, determining the subset of initial features may be referred to as feature selection. In some cases, the selected features may be expected to contain relevant information from the input data (e.g., EEG signals or temporal segments of EEG signals). In some cases, the selected features may be expected to contain relevant information from the input data such that a desired task can be performed by using this reduced representation instead of the complete initial data.

[0121] In some embodiments, feature extraction may involve reducing the number of resources required to describe a large set of data (e.g., an EEG signal or a time segment of an EEG signal). In some cases, analysis involving a large number of variables may require large amounts of memory and computational power. In some cases, feature extraction may construct combinations of variables to accurately describe the data with sufficient accuracy. In some cases, feature extraction may construct combinations of variables to accurately describe the data with sufficient accuracy while preventing overfitting.

[0122] In some embodiments, the results may be improved using a constructed set of application-dependent features. In some cases, the constructed set may be built by experts. In some cases, general dimensionality reduction techniques may be used. In some cases, the general dimensionality reduction techniques may be, for example, independent component analysis, isomap, kernel PCA, latent semantic analysis, partial least squares, principal component analysis, multi-factor dimensionality reduction, nonlinear dimensionality reduction, multilinear principal component analysis, multilinear subspace learning, semidefinite embedding, autoencoder, etc.

[0123] In some cases, the set of numerical features may be described by a feature vector, which may be an n-dimensional vector of numerical features that represents an object, as an example an EEG signal or a temporal segment of an EEG signal.

[0124] In some embodiments, a data analysis software package may provide feature extraction. In some cases, the data analysis software package may provide dimensionality reduction. In some cases, the data analysis software package may include a programming environment such as MATLAB, SciLab, NumPy, or R language. In some cases, a programming language script may be used to extract features from the EEG signal. In some cases, the programming language script may be, for example, MATLAB, Python, Java, JavaScript, Ruby, C, C++, or Perl.

[0125] In some cases, the features may be unique within the EEG signals or a temporal segment thereof. Unique may be a feature of the EEG signal that may be measured, such as the amplitude of the EEG signal, the duration of the EEG signal, the variance of the EEG signal, the power of the EEG signal, local maxima / minima of the EEG signal, the pattern of the EEG signal, the regularity of the EEG signal, the spectral power distribution of the EEG signal, or the frequency of the EEG signal. In some cases, the features may be a measurement of the power of the signal within a particular frequency band. The frequency band may be, for example, about 0 Hz to 100 Hz. In some cases, the power of the signal may be normalized to the total power. In some cases, the power of the signal may be a ratio of the power between one or more frequency bands. In some cases, the feature may be a function performed on the signal to obtain a value. For example, the function may measure the root mean square (RMS) of the signal (e.g., EEG signal) to obtain the RMS value of the signal. In some cases, the feature may compare one signal (e.g., EEG signal) to one or more signals. In some cases, the features may compare one or more signals (e.g., EEG signals) to one or more signals. In some cases, the features may measure attributes of the signals (e.g., EEG signals). In some cases, the features may compare one or more attributes of the signals (e.g., EEG signals) to one or more attributes of the signals. The attributes may be, for example, inherent properties of the EEG signals. In some cases, the features of the EEG signals may be continuous and / or discrete time.

[0126] In some cases, the plurality of features may include at least 20 different time and / or frequency features. In some cases, the plurality of features may include up to 1,000 time and / or frequency features. In some cases, the plurality of features may include between about 10 features and 200 features. In some cases, the plurality of features may include between about 10 features and 100 features. In some cases, the plurality of features may include between about 10 features and 50 features.

[0127] In some cases, the features may include multiple discrete values ​​associated with time domain, frequency domain, time-frequency domain, information theory, and nonlinear dynamical systems theory features. In some cases, the features may include multiple discrete values ​​associated with time and / or frequency domain features. The features may include multiple continuous values ​​associated with time and / or frequency domain features or morphological structure of the signal. In some cases, the features may also be brain asymmetry, amplitude variation, spatial and temporal correlation, coherence, or covariates of two or more features. In some cases, the features may also be frequency spectrum and characteristics of the EEG signal. In some cases, the signal characteristics may include jitter, distortion, spread spectrum, time measurements, frequency measurements, etc. In some cases, analyzing the features may include ranking and / or classifying the features.

[0128] In some cases, the signals may be converted to digital signals. In some cases, the signals may be converted to digital signals and then to analog signals.

[0129] In some cases, the features may be sampled from a portion of the EEG signal.The features may be sampled from a portion of the EEG signal to reduce the processing time and power required.

[0130] In some embodiments, the features may be associated with a weight value. The weight value may give a higher score to one feature to detect a neurological condition or a particular neurological condition. A higher score may indicate that the feature may be more relevant in predicting one or more neurological conditions. The score may indicate that the feature may be more relevant in predicting a neurological condition. The score may indicate that the feature may be more relevant in predicting the severity of a particular neurological condition. The method may adjust the weight value of any feature at any given time. The method may adjust the weight value of any feature at any given time by increasing and / or decreasing.

[0131] In a variation in which delirium is to be detected, features may be calculated using time-frequency analysis of the data recorded on each channel for each time window, for example, for each 4-second window, for each 10-second window, for each 15-second window, for each 30-second window, or for each 60-second window, to produce single-channel features. In addition, a set of features may be calculated to analyze the signal interactions between pairs of channels, called multi-channel features. Exemplary single-channel features may include power, spectral properties, power ratios, amplitude characteristics and morphological structure features, entropy, variability, and wavelet decomposition in different frequency bands (e.g., alpha, beta, delta, theta, and gamma), and correlations within and across hemispheres for different frequency bands (e.g., alpha, beta, delta, theta, and gamma), as well as spectral, amplitude, and phase associated correlations, and synchrony measures.

[0132] An EEG window may be marked as an artifact if a set of predefined features exceeds a certain threshold. Additionally, a segment may also be marked as an artifact if the most recently reported impedance on an electrode is higher than a pre-set threshold. When an EEG window is marked as an artifact, it is generally not used for subsequent analysis and prediction.

[0133] Classification using machine learning In some embodiments, the method may include applying a machine learning algorithm to the features to perform classification for one or more neurological conditions for each temporal data segment for each channel individually. In some cases, the machine learning classification module may include performing classification for one or more neurological conditions for each temporal data segment for all channels together. In some cases, the machine learning classification module may include performing classification for one or more neurological conditions for each temporal data segment of one or more groups, each group consisting of one or more channels. Figure 2 shows a machine learning classification module 250 that may determine the collected / extracted features from the preprocessing step and classify the features. In some cases, the features may be extracted without a preprocessing step.

[0134] In some cases, machine learning algorithms may need to extract and draw relationships between features because traditional statistical techniques may not be sufficient. In some cases, machine learning algorithms may be used in conjunction with traditional statistical techniques. In some cases, traditional statistical techniques may provide pre-processed features to the machine learning algorithm.

[0135] In some embodiments, multiple features may be used by one or more machine learning algorithms to provide a classification over a time segment for neurological conditions.

[0136] In some embodiments, a cluster of positive neurological indications may consist of between about 1 and 50 positive neurological indications. In some cases, a cluster of positive classifications may consist of between 1 and 10 positive neurological indications.

[0137] In some embodiments, for each corresponding time epoch, the cluster of delirium positive classifications may indicate various levels of delirium and / or delirium. In some cases, the cluster of delirium indication positive classifications may comprise between about 1 and 50 delirium indication positive classifications. In some cases, the cluster of positive classifications may comprise between 1 and 10 delirium indication positive classifications.

[0138] In some embodiments, for each corresponding time epoch, the clusters of sedation positive classifications may indicate various levels of sedation and / or sedation. In some cases, the clusters of delirium indication positive classifications may comprise between about 1 and 50 delirium indication positive classifications. In some cases, the clusters of positive classifications may comprise between 1 and 10 delirium indication positive classifications.

[0139] In some embodiments, for each corresponding time epoch, the clusters of seizure positive classifications may indicate different levels of seizures and / or seizures. In some cases, the clusters of seizure indication positive classifications may consist of about 1-50 seizure indication positive classifications. In some cases, the clusters of positive classifications may consist of 1-10 seizure indication positive classifications.

[0140] In some embodiments, for each corresponding time epoch, the cluster of stroke positive classifications may indicate various levels of stroke and / or stroke. In some cases, the cluster of stroke indication positive classifications may consist of about 1-50 stroke indication positive classifications. In some cases, the cluster of positive classifications may consist of 1-10 stroke indication positive calculations.

[0141] In some embodiments, the method may further include comparing the classifications sequentially across multiple time epochs on each channel. In some cases, before / after / during comparing the classifications sequentially across multiple time epochs on each channel, the sequential classifications across multiple time epochs on each channel may be discarded. In some cases, a subset of the classifications may be discarded. In some cases, a subset of less than about 1-20 classifications may be discarded. In some cases, a subset of less than 3 classifications may be discarded. In some cases, a subset of less than 7 classifications may be discarded. In some cases, a subset of less than 10 classifications may be discarded. In some cases, a subset of less than 15 classifications may be discarded. In some cases, a subset of less than 20 classifications may be discarded.

[0142] In some embodiments, a subset of the neurological indication positive classifications may be discarded, for example because they may be random readings such as low reliability, inaccurate classifications, incorrect classifications, calibration, system errors, disconnected electrodes, artifact signals, system interference, or other signals.

[0143] In some embodiments, a subset of the neurological indication positive classifications may be discarded, for example, to save memory space, improve processing speed, reduce energy usage, reduce system heat, reduce computational costs, save processing power, save processing time, increase reliability, or reduce random access memory usage, etc.

[0144] In some embodiments, a higher number of consecutive positive neurological indications classifications may indicate higher reliability.The higher the reliability of positive neurological indications classifications, the more accurate the determination of the detection of one or more neurological conditions in the patient.In some cases, the higher reliability of positive neurological indications classifications may indicate machine learning algorithm accuracy, data (EEG signal) quality, or EEG detection system health status, etc.

[0145] In some embodiments, a particular time epoch may be classified as associated with one or more neurological conditions if a temporal data segment for a subset of the plurality of channels is classified as positive for a neurological indication. In some cases, the subset may be at least 5%, 10%, 20%, 30%, 40%, 50%, or more of the plurality of channels. In some cases, the subset may be up to about 50%, 40%, 30%, 20%, 10%, 5%, or less of the plurality of channels.

[0146] In some embodiments, a particular time epoch may be classified as associated with a sedated state if a temporal data segment for a subset of the plurality of channels is classified as a positive sedation state indication. In some cases, the subset may be at least 5%, 10%, 20%, 30%, 40%, 50%, or more of the plurality of channels. In some cases, the subset may be up to about 50%, 40%, 30%, 20%, 10%, 5%, or less of the plurality of channels.

[0147] In some embodiments, a particular time epoch may be classified as associated with delirium if a temporal data segment for a subset of the plurality of channels is classified as positive for delirium indication. In some cases, the subset may be at least 5%, 10%, 20%, 30%, 40%, 50%, or more of the plurality of channels. In some cases, the subset may be up to about 50%, 40%, 30%, 20%, 10%, 5%, or less of the plurality of channels.

[0148] In some embodiments, a particular time epoch may be classified as associated with a stroke if a temporal data segment for a subset of the plurality of channels is classified as stroke indication positive. In some cases, the subset may be at least 5%, 10%, 20%, 30%, 40%, 50%, or more of the plurality of channels. In some cases, the subset may be up to about 50%, 40%, 30%, 20%, 10%, 5%, or less of the plurality of channels.

[0149] In some embodiments, a particular time epoch may be classified as associated with a seizure if a temporal data segment for a subset of the plurality of channels is classified as seizure indication positive. In some cases, the subset may be at least 5%, 10%, 20%, 30%, 40%, 50%, or more of the plurality of channels. In some cases, the subset may be up to about 50%, 40%, 30%, 20%, 10%, 5%, or less of the plurality of channels.

[0150] In some embodiments, the classification may include assigning a probability value, for example, 0-1 or 0-100, to the time segment reflecting that the subject has a neurological condition, e.g., sedation, delirium, stroke, seizure, etc. In some embodiments, the classification may include assigning a severity value, for example, 0-7, to the time segment reflecting the severity of the neurological condition, e.g., sedation, delirium, stroke, seizure, etc. In some embodiments, the value assigned in the classification is a combined value reflecting both probability and severity.

[0151] In some embodiments, the classification may be a binary classification, such as neurological condition positive and neurological condition negative. Alternatively, the score may be a classification of the temporal segment into one of three or more categories for a neurological condition. The temporal segment may be classified as, by way of example, neurological condition positive, neurological condition negative, neurological condition-like, uncertain neurological condition activity, non-neurological condition activity, artifact, etc. The neurological condition classification may relate, for example, to sedation, delirium, stroke, seizure, etc. The temporal segment may be classified as, for example, sedation positive, sedation negative, sedation-like, uncertain sedation activity, non-neurological condition activity, delirium positive, delirium negative, delirium-like, uncertain delirium activity, non-delirium activity, stroke positive, stroke negative, stroke-like, uncertain stroke activity, non-stroke activity, seizure positive, seizure negative, seizure-like, uncertain seizure activity, non-seizure activity, artifact, etc. The temporal segment may be classified, for example, as sedation-like and delirium-like. The temporal segments may be classified, for example, as sedation positive, sedation negative, sedation-like, uncertain sedation activity, delirium positive, delirium negative, delirium-like, uncertain delirium activity, and non-delirium activity.

[0152] The neurological condition classification may be associated with one or more different neurological conditions. In some cases, the features may be classified into 1-20 categories. In some cases, the features may be classified into 1-10 categories. Each category may also be divided into subcategories. For example, a neurological condition may be divided into one or more conditions associated with, for example, sedation, delirium, stroke, seizures, etc. In some cases, a neurological condition may be divided into one or more subcategories associated with a particular concomitant assessment score. For example, a sedation positive score may be classified as a value on the RASS scale, the SAS scale, and / or the RSS scale. For example, a delirium positive score may be classified as a value on the CAM-ICU or CAM-ICU-7 scale. The values ​​may be used to determine the features associated with a particular score on one or more concomitant assessment scores. In another example, a time segment classified as stroke positive may be further subdivided into classifications based on the type of stroke. In some cases, the stroke positive classification may be subdivided by stroke type, stroke location or hemisphere, or stroke size or severity. Large vessel occlusion (LVO) strokes may be subdivided into classifications based on the vessel in which the occlusion resides.

[0153] In some embodiments, the method may include applying a machine learning algorithm to the plurality of features to perform a neurological multi-class classification (neurological condition positive, neurological condition negative, neurological condition like, uncertain neurological condition activity, non-neurological condition activity, etc.) per channel and per temporal segment.

[0154] In some embodiments, the method may include applying a machine learning algorithm to the multiple features to perform neurological multi-class classification per channel and per temporal segment (e.g., sedated vs. delirium vs. stroke vs. seizures, sedated vs. delirium vs. stroke, delirium vs. stroke vs. seizures, sedated vs. delirium, sedated vs. stroke, sedated vs. seizures, delirium vs. stroke, delirium vs. stroke, stroke vs. seizures, stroke vs. non-stroke, LVO vs. non-LVO, etc.).

[0155] In some embodiments, the method may include applying a machine learning algorithm to the features to perform a binary neurological classification (e.g., neurological condition positive vs. neurological condition negative) for each channel and for each temporal data segment. In some embodiments, one or more collected features may be discarded prior to or during the machine learning classification or prior to categorization.

[0156] In some embodiments, the method may include applying a machine learning algorithm to the features to perform a multi-class stroke classification (stroke positive, stroke negative, stroke-like, uncertain stroke activity, non-stroke activity, etc.) per channel and per temporal segment. In some embodiments, the method may include applying a machine learning algorithm to the features to perform a binary stroke classification (e.g., stroke vs. non-stroke, LVO vs. non-LVO) per channel and per temporal data segment. In some embodiments, one or more collected features may be discarded prior to or during the machine learning classification or prior to categorization.

[0157] In some embodiments, the method may include applying a machine learning algorithm to a plurality of features to perform a delirium multi-class classification (e.g., delirium positive, delirium negative, delirium-like, uncertain delirium activity, etc.) per channel and per temporal segment. In some embodiments, the method may include applying a machine learning algorithm to a plurality of features to perform a delirium binary classification (e.g., delirium vs. not delirium) per channel and per temporal data segment.

[0158] In some embodiments, the method may include applying a machine learning algorithm to a plurality of features to perform a sedation state multi-class classification (sedated positive, sedated negative, sedated-like, uncertain sedated activity, etc.) per channel and per temporal segment. In some embodiments, the method may include applying a machine learning algorithm to a plurality of features to perform a sedation state binary classification (e.g., sedated vs. non-sedated) per channel and per temporal data segment.

[0159] In some embodiments, the method may include applying a machine learning algorithm to a plurality of features to perform a seizure multi-class classification (e.g., seizure positive, seizure negative, seizure-like, uncertain seizure activity, etc.) per channel and per temporal segment. In some embodiments, the method may include applying a machine learning algorithm to a plurality of features to perform a sedation state binary classification (e.g., seizure vs. not seizure) per channel and per temporal data segment.

[0160] In some embodiments, a human may select and discard features prior to / during machine learning classification. In some cases, a computer may select and discard features. In some cases, features may be discarded based on a threshold. In some cases, features may be discarded based on one or more contingent assessment tests and / or specific values ​​within the contingent assessment tests.

[0161] In some embodiments, any number of features may be classified by the machine learning algorithm. The machine learning algorithm may classify at least 10 features. In some cases, the plurality of features may include between about 10 features and 1,000 features. In some cases, the plurality of features may include between about 10 features and 200 features. In some cases, the plurality of features may include between about 10 features and 100 features. In some cases, the plurality of features may include between about 10 features and 50 features. In some embodiments, the machine learning algorithm may be, for example, an unsupervised learning algorithm, a supervised learning algorithm, or a combination thereof. The unsupervised learning algorithm may be, for example, clustering, hierarchical clustering, k-means, mixture models, DBSCAN, OPTICS algorithm, anomaly detection, local outlier factor methods, neural networks, autoencoders, deep belief networks, Hebbian learning, generative adversarial networks, self-organizing maps, expectation maximization algorithm (EM), method of moments, blind source separation techniques, principal component analysis, independent component analysis, non-negative matrix factorization, singular value decomposition, or a combination thereof. The supervised learning algorithm may be, for example, a support vector machine, a linear regression, a logistic regression, a linear discriminant analysis, a decision tree, a k-nearest neighbor algorithm, a neural network, similarity learning, or a combination thereof. In some embodiments, the machine learning algorithm may include a deep neural network (DNN). The deep neural network may include a convolutional neural network (CNN). The CNN may be, for example, U-Net, ImageNet, LeNet-5, AlexNet, ZFNet, GoogleNet, VGGNet, ResNet18, or ResNet, etc.Other neural networks may be, for example, deep feedforward neural networks, recurrent neural networks, LSTMs (long short-term memories), GRUs (gated recurrent units), autoencoders, variational autoencoders, adversarial autoencoders, denoising autoencoders, sparse autoencoders, restricted Boltzmann machines, RBMs (restricted BMs), deep belief networks, generative adversarial networks (GANs), deep residual networks, capsule networks, or attention / transformer networks, etc.

[0162] In some embodiments, the machine learning algorithm may be, for example, a naive Bayes classifier, linear regression, logistic regression, decision tree, random forest, rotation forest, K-nearest neighbors (KNN), clustering, support vector machine (SVM), or neural network. In some cases, the machine learning algorithm may include ensemble algorithms such as bagging, boosting, and stacking. The machine learning algorithm may be applied to multiple features individually, extracted per channel, so that each channel may have a separate iteration of the machine learning algorithm, or applied to multiple features at once, extracted from all channels or a subset of channels.

[0163] In some embodiments, the method may apply one or more machine learning algorithms. In some embodiments, the method may apply one or more machine learning algorithms per channel.

[0164] In FIG. 2, the machine learning classification module 250 may include any number of machine learning algorithms. In some embodiments, the random forest machine learning algorithm may be an ensemble of bagged decision trees. In some cases, the ensemble of bagged decision trees may classify each temporal data segment per channel as (1) a neurological condition positive or (2) a neurological condition negative. In some cases, the ensemble of bagged decision trees may classify each temporal data segment per channel as (1) a delirium positive or (2) a delirium negative. In some cases, the ensemble of bagged decision trees may classify each temporal data segment per channel as (1) a sedation positive or (2) a sedation negative. In some cases, the ensemble of bagged decision trees may classify each temporal data segment per channel as (1) a stroke positive or (2) a stroke negative. In some cases, the ensemble of bagged decision trees may classify each temporal data segment per channel as (1) a seizure positive or (2) a seizure negative.

[0165] The ensemble may be at least about 1, 2, 3, 4, 5, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 120, 140, 160, 180, 200, 250, 500, 1,000 or more bagged decision trees. The ensemble may be up to about 1,000, 500, 250, 200, 180, 160, 140, 120, 100, 90, 80, 70, 60, 50, 40, 30, 20, 10, 5, 4, 3, 2 or less bagged decision trees. The ensemble may be about 1-1,000, 1-500, 1-200, 1-100, or 1-10 bagged decision trees.

[0166] In some embodiments, the method may include applying the machine learning classifier to any number of channels. The method may include applying the machine learning classifier to at least about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20, 25, 50, 100, 500, 1,000, or more channels. The method may include applying the machine learning classifier to up to about 1,000, 500, 100, 50, 25, 20, 15, 10, 9, 8, 7, 6, 5, 4, 3, 2, or less channels. The method may include applying the machine learning classifier to about 1-1,000, 1-100, 1-25, or 1-5 channels.

[0167] In some cases, multiple EEG signals may be collected across multiple channels. Machine learning algorithms may be applied to features extracted per channel individually, such that each channel has a separate iteration of the machine learning algorithm, or to features extracted from all channels or a subset of channels at once. Each channel may have at least about 1, 2, 5, 10, 25, 50, or more machine learning algorithms applied. Each channel may have up to about 50, 25, 10, 5, 2, or fewer machine learning algorithms applied.

[0168] In some embodiments, the method may include applying a machine learning classifier to a subset of channels. The subset of channels may be at least about 1%, 5%, 10%, 20%, 30%, 40%, 50%, or more of the total set of channels. The subset of channels may be up to about 50%, 40%, 30%, 20%, 10%, 5%, 1%, or less of the total set of channels. The subset of channels may be about 1%-50%, 1%-40%, 1%-30%, 1%-20%, 1%-10%, or 1%-5% of the total set of channels.

[0169] In some embodiments, the machine learning algorithm may have various parameters, such as a learning rate, a small batch size, a number of epochs to train, momentum, a learning weight decay rate, or neural network layers.

[0170] In some embodiments, the learning rate may be between about 0.00001 and 0.1.

[0171] In some embodiments, the small batch size may be about 16-128.

[0172] In some embodiments, the neural network may comprise neural network layers. The neural network may have at least about 2 to 1,000 or more neural network layers.

[0173] In some embodiments, the number of epochs for training may be at least about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 25, 30, 35, 40, 45, 50, 55, 60, 65, 70, 75, 80, 85, 90, 95, 100, 150, 200, 250, 500, 1,000, 10,000, or more.

[0174] In some embodiments, the momentum may be at least about 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, or more, In some embodiments, the momentum may be up to about 0.9, 0.8, 0.7, 0.6, 0.5, 0.4, 0.3, 0.2, 0.1, or less.

[0175] In some embodiments, the learning weighted decay rate may be at least about 0.00001, 0.0001, 0.001, 0.002, 0.003, 0.004, 0.005, 0.006, 0.007, 0.008, 0.009, 0.01, 0.02, 0.03, 0.04, 0.05, 0.06, 0.07, 0.08, 0.09, 0.1, or more. In some embodiments, the learning weighted decay rate may be up to about 0.1, 0.09, 0.08, 0.07, 0.06, 0.05, 0.04, 0.03, 0.02, 0.01, 0.009, 0.008, 0.007, 0.006, 0.005, 0.004, 0.003, 0.002, 0.001, 0.0,001, 0.00,001, or less.

[0176] In some embodiments, the machine learning algorithm may use a loss function, which may be, for example, a regression loss, a mean absolute error, a mean bias error, a hinge loss, an Adam optimizer, and / or a cross-entropy.

[0177] In some embodiments, the parameters of the machine learning algorithm may be adjusted with the aid of a human and / or a computer system.

[0178] In some embodiments, the machine learning algorithm may prioritize certain features. The machine learning algorithm may prioritize features that may be more relevant for detecting one or more neurological conditions, a particular neurological condition, a condition of a subject associated with a neurological condition, delirium, seizures, sedation, stroke, and / or any combination thereof. A feature may be more relevant for detecting one or more neurological conditions if the feature is classified more frequently than another feature. A feature may be more relevant for detecting delirium, seizures, sedation, stroke, and / or any combination thereof if the feature is classified more frequently than another feature. In some cases, the features may be prioritized using a weighting system. In some cases, the features may be prioritized on a probability statistic based on the frequency and / or number of occurrences of the feature. The machine learning algorithm may prioritize features with the aid of a human and / or a computer system.

[0179] In some embodiments, one or more of the features may be used in conjunction with machine learning or traditional statistical techniques to determine whether a segment is likely to contain an artifact. FIG. 2 shows an artifact removal module 255, which identifies segments that contain artifacts. The identified artifacts may be the result of electrical interference, electrode instability or movement, subject movement, subject eye movement or blinking, subject chewing, subject muscle tension, subject electrocardiogram artifacts, etc. In some cases, a movement sensor or other sensor may be used as an additional input to the artifact removal module. In some cases, the identified artifacts may be removed so that they are not used in stroke classification. In some cases, the identified artifacts may be reduced, eliminated, or eliminated, and the remaining signal may still be processed for stroke classification.

[0180] In some cases, a machine learning algorithm may prioritize certain features to reduce computational cost, save processing power, save processing time, increase reliability, reduce random access memory usage, etc.

[0181] III. Neurological Conditional Probability / Classification and Output Control policies and neurological conditional probability / classification In some embodiments, the multi-neurological classification may include classifying, for each channel, each temporal data segment as (1) a neurological condition positive or (2) a neurological condition negative. In some embodiments, the multi-neurological classification may include classifying, for each channel, each temporal data segment as (1) a sedation positive or (2) a sedation negative. In some embodiments, the multi-neurological classification may include classifying, for each channel, each temporal data segment as (1) a delirium positive or (2) a delirium negative. In some embodiments, the multi-neurological classification may include classifying, for each channel, each temporal data segment as (1) a seizure positive or (2) a seizure negative. In some embodiments, the multi-neurological classification may include classifying, for each channel, each temporal data segment as (1) a stroke positive or (2) a stroke negative.

[0182] The multi-neurological classification may use a machine learning algorithm, as described elsewhere herein. The method may include aggregating the multi-neurological classification across multiple temporal data segments for multiple channels over a moving time window. The method may include aggregating the sedation and delirium classification across multiple temporal data segments for multiple channels over a moving time window. The classification may be used to determine the patient's level of sedation and / or delirium on a particular scale or assessment test.

[0183] The method may include aggregating multiple delirium classifications and multiple sedation classifications across multiple temporal data segments for multiple channels over a moving time window. The method may include aggregating multiple sedation classifications across multiple temporal data segments for multiple channels over a moving time window. The method may include aggregating multiple delirium classifications across multiple temporal data segments for multiple channels over a moving time window. The method may include aggregating multiple stroke classifications across multiple temporal data segments for multiple channels over a moving time window. The method may include aggregating multiple seizure classifications across multiple temporal data segments for multiple channels over a moving time window. The method may include aggregating multiple seizure classifications across multiple temporal data segments for multiple channels over a moving time window to aid in determining a patient's level of sedation and / or delirium on a particular scale or assessment test.

[0184] The aggregated neurological classification may follow a control policy module 275 of the neurological condition probability / classification calculation and output module 270, as shown in FIG. 2. FIG. 2 illustrates the neurological condition probability / classification calculation and output module 270. As shown in FIG. 2, the neurological condition probability / classification calculation and output module 270 may comprise a control policy module 275, a neurological condition probability / classification calculation module 280, a neurological condition probability plot module 285, and a neurological condition notification module 290. The control policy module 275 may be configured to implement a control policy, each as described elsewhere herein, the neurological condition probability / classification calculation module 280 may be configured to calculate a neurological condition probability, the neurological condition probability plot module 285 may be configured to plot a neurological condition probability value, and the neurological condition notification module 290 may be configured to provide a notification and / or assessment.

[0185] The processing module 120 may further be configured to generate a visual output. The visual output may include a graph displaying the probability / severity that the subject is suffering from one or more neurological conditions. The graphical representation may be a combination of multiple different temporal graphs corresponding to multiple neurological conditions. The graphical representation may include an overlay of multiple different temporal graphs. As shown in FIG. 4, the graphical representation 400 may include seizure 440, stroke 430, sedation 420, and delirium 410.

[0186] The processing module 120 may be configured to generate one or more contingent assessment scores indicative of the severity of the one or more neurological conditions. The processing module 120 may be configured to generate a diagnostic output 290 based on the indication or assessment. The diagnostic output may include an aggregate wellness score or a graphical representation of the subject's brain state. The aggregate wellness score may be a combination of multiple discrete scores corresponding to multiple neurological conditions. The multiple discrete scores may be combined based on different weights allocated to the multiple neurological conditions. The processing module 120 may be configured to generate one or more contingent assessment scores indicative of the one or more neurological conditions. The method may include generating one or more notifications when the patient has a wellness score below or above a particular wellness score.

[0187] In some cases, the moving window may have a time period of 1 minute to 1 hour. In some cases, the time period of the moving window may be dynamic or adjustable instead of fixed. In some cases, the time period of the moving window may be subject dependent.

[0188] In some embodiments, a cluster of neurological condition positive classifications on one or more channels may result in an overall determination of the neurological condition for the patient over the corresponding time epoch in accordance with the control strategy module 275. In some embodiments, a cluster of sedation positive and / or delirium positive classifications on one or more channels may result in an overall determination of the sedation state and / or delirium level for the patient over the corresponding time epoch in accordance with the control strategy module 275.

[0189] The control policy may be a set of rules that results in an overall determination of a neurological condition diagnosis or probability for the patient. The control policy may be a set of rules that results in an overall determination of a delirium and / or sedation diagnosis or probability for the patient. The control policy may take a set of parameters as input and act on the set of parameters according to a set of rules to result in an overall determination of a neurological condition for the patient. The control policy may take a set of parameters as input and act on the set of parameters according to a set of rules to result in an overall determination of a sedation state and / or delirium level for the patient. The set of rules may be as described anywhere herein. The set of rules may be adjusted at any time to act on more parameters or act on fewer parameters. The set of rules may be adjusted at any time to include more rules or remove rules. The set of rules may be at least about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20, 25, 50, 100, 500, 1,000, or more rules. The set of rules may be up to about 1,000, 500, 100, 50, 25, 20, 15, 10, 9, 8, 7, 6, 5, 4, 3, 2, or less rules. The set of rules may be about 1-1,000, 1-500, 1-100, 1-25, 1-10, 1-5, or 1-3 rules.

[0190] In some embodiments, parameter inputs for the control strategy may include the number of classifications of channels as positive for a neurological condition, the number of classifications of channels as negative for a neurological condition, classification of channels as positive for a neurological condition, classification of channels as negative for a neurological condition, classification of channels as neurological conditions, corresponding time epochs, number of channels, the machine learning algorithm used for classification, moving window time length, quality of connection for each channel, information derived from the EKG signal, information derived from the EMG signal, information regarding patient demographics, information regarding the patient's current or previous condition, information regarding a treatment or medication regimen applied to the patient, information derived from a movement sensor (e.g., an accelerometer or inertial measurement unit), etc.

[0191] In some embodiments, a control strategy may have inputs of any number of parameters. A control strategy may have inputs of at least about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 20, 25, 50, 100, 500, 1,000, or more parameters. A control strategy may have inputs of up to about 1,000, 500, 100, 50, 25, 20, 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, or less parameters. A control strategy may have inputs of about 1-1,000, 1-500, 1-100, 1-50, 1-25, 1-15, 1-10, or 1-5 parameters.

[0192] In some embodiments, the set of rules may dictate that the control policy discard the classification of the channel. For example, if the control policy receives an input of a single neurological condition positive classification over the corresponding time epoch, the set of rules may discard the neurological condition positive classification over the corresponding time epoch. For example, if the control policy receives an input of a single sedation positive and / or delirium positive classification over the corresponding time epoch, the set of rules may discard the delirium positive and / or classification over the corresponding time epoch.

[0193] In some cases, the control policy may receive at least about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 50, 100, 500, or more positive classifications, and the set of rules may discard each positive classification over the corresponding time epoch. In some cases, the control policy may receive at most about 500, 100, 50, 10, 9, 8, 7, 6, 5, 4, 3, 2, or less positive classifications of strokes, and the set of rules may discard each positive classification of strokes over the corresponding time epoch. In some cases, the control policy may receive about 1-500, 1-100, 1-50, 1-10, or 1-5 positive classifications, and the set of rules may discard each positive classification of strokes over the corresponding time epoch. The positive classification may be associated with a neurological condition. The positive classification may be associated with a sedated state and / or delirium. The positive classification may be associated with a seizure. The positive classification may be associated with a stroke.

[0194] In some embodiments, the set of rules may dictate that the control policy outputs a neurological condition positive classification for a set of channels corresponding to a time epoch. For example, if the control policy receives a set of four or more channels, each of which registers a neurological condition positive classification over a corresponding time epoch, the set of rules may output a neurological condition positive classification over the corresponding time epoch. In some cases, the control policy may receive a set of neurological condition positive classifications of at least about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20, 25, 50, 100, 1,000, or more channels, and the set of rules may output a neurological condition positive classification for the set of stroke positive classifications over the corresponding time epoch. In some cases, the control strategy may receive a set of neurological condition positive classifications of up to about 1,000, 100, 50, 25, 20, 15, 10, 9, 8, 7, 6, 5, 4, 3, 2, or less channels, and the set of rules may output a neurological condition positive classification for the set of neurological condition positive classifications over a corresponding time epoch. In some cases, the control strategy may receive a set of about 1-1,000, 1-500, 1-100, 1-50, 1-25, 1-10, or 1-5 neurological condition positive classifications, and the set of rules may output a neurological condition positive classification for the set of neurological condition positive classifications over a corresponding time epoch.

[0195] In some embodiments, the set of rules may dictate that the control policy outputs a delirium positive and / or sedation positive classification for a set of channels corresponding to a time epoch. For example, if the control policy receives a set of four or more channels that each register a delirium positive and / or sedation positive classification over a corresponding time epoch, the set of rules may output a stroke positive classification over the corresponding time epoch. In some cases, the control policy may receive a set of delirium positive and / or sedation positive classifications for at least about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20, 25, 50, 100, 1,000, or more channels, and the set of rules may output a delirium positive and / or sedation positive classification for the set of delirium positive and / or sedation positive classifications over the corresponding time epoch. In some cases, the control policy may receive a set of delirium positive and / or sedation positive classifications for up to about 1,000, 100, 50, 25, 20, 15, 10, 9, 8, 7, 6, 5, 4, 3, 2, or less channels, and the set of rules may output a delirium positive and / or sedation positive classification for the set of delirium positive and / or sedation positive classifications over a corresponding time epoch. In some cases, the control policy may receive a set of about 1-1,000, 1-500, 1-100, 1-50, 1-25, 1-10, or 1-5 delirium positive and / or sedation positive classifications, and the set of rules may output a delirium positive and / or sedation positive classification for the set of delirium positive and / or sedation positive classifications over a corresponding time epoch.

[0196] In some embodiments, the method may include calculating a neurological condition probability / classification for the patient as a percentage of neurological condition positive classifications within a defined time period. In some embodiments, the method may include calculating a delirium and / or sedation state probability / classification for the patient as a percentage of delirium positive and / or delirium negative classifications within a defined time period.

[0197] As shown in FIG. 2, the neurological condition probability / classification calculation module 280 may be configured to calculate a neurological condition probability of the patient. The neurological condition probability / classification calculation module 280 may be configured to classify a neurological condition (e.g., delirium, sedation, seizure, stroke, etc.) of the patient. In some cases, the time period used for the neurological condition probability / classification calculation may be from 1 minute to 1 hour. In some cases, the time period used for the neurological condition probability / classification calculation may be the entirety of the recording session. In some cases, the time period used for the neurological condition probability / classification calculation may be dynamic or adjustable instead of being fixed. In some cases, the time period used for the sedation and / or delirium probability / classification calculation may be from 1 minute to 1 hour. In some cases, the time period used for the sedation and / or delirium probability / classification calculation may be the entirety of the recording session.

[0198] In some embodiments, the neurological condition probability may form a continuous output that is measured by calculating the neurological condition probability over a moving window of time, resulting in a neurological condition probability / classification calculation over each sequential time period. In some embodiments, the sedation and / or delirium probability may form a continuous output that is measured by calculating the sedation and / or delirium probability over a moving window of time, resulting in a sedation and / or delirium probability / classification calculation over each sequential time period. In some cases, the time period of the moving window may be from 1 minute to 1 hour. In some cases, the time period of the moving window may be dynamic or adjustable instead of fixed. In some cases, the sequential time periods formed by the moving window may overlap. In some cases, the sequential time periods formed by the moving window may be non-overlapping. In some cases, the moving window may move in time increments from 1 second to 1 hour. In some cases, the moving window may pause or skip time periods such that the resulting neurological condition probability / classification calculations are not continuous or sequential in time.

[0199] FIG. 2 illustrates a neurological condition probability plot module 285 configured to plot the probability of a neurological condition of a subject. As illustrated in FIG. 3, a neurological condition probability output 315 may be displayed to a user via an interface 300. The severity of one or more neurological conditions 320 may also be displayed to a user. In some embodiments, the neurological condition probability output may display one or more adjustable thresholds to a user on a time series plot. In some embodiments, the neurological condition probability output may be displayed to a user as a time series plot, a bar graph, a chart, or the like. As illustrated in FIG. 4, one or more neurological conditions may be illustrated as a time vs. probability 400. FIG. 4 illustrates a patient's minute-by-minute probability percentages over time for seizure 440, stroke 430, sedation 420, and delirium 410. The time series plot may plot the probability for seizure, stroke, sedation, delirium, or any combination thereof (e.g., seizure and delirium, sedation and delirium, stroke and sedation and delirium, etc.). As shown in FIG. 5, one or more neurological conditions may be illustrated individually or with each other. A semicircular meter plot may indicate the severity or probability of a neurological condition. The further to the right the indication line is, the higher the severity or probability of the neurological condition. The further to the left the indication line is, the less severe or less likely the neurological condition is. As shown in FIG. 5, the system may include semicircular meter plots for seizure 510, stroke 530, delirium 520, and sedation 540.

[0200] In some embodiments, the time series plot may be depicted in a color to highlight when a threshold has been crossed or a particular severity has been reached. For example, if the probability of delirium exceeds a delirium threshold probability / severity value, a notification may be sent by the system. In some cases, the neurological condition probability plot module may display a variety of information, such as the time period measured, the date, or the time of initial acquisition. In some cases, the neurological condition probability plot may be usable by a health care practitioner to assess the subject's condition and determine a course of treatment. The neurological condition probability plot may also be usable by a health care practitioner to monitor the progression of the subject's condition over time or to monitor the effectiveness of a course of treatment.

[0201] FIG. 2 illustrates a notification module 290 configured to generate notifications. In some embodiments, the method may include generating one or more notifications when a neurological condition classification is made. In some embodiments, the method may include generating one or more notifications when a sedation, delirium, seizure, and / or stroke classification is made. In some embodiments, the method may include generating one or more notifications when a neurological condition positive classification is made or when a neurological condition probability equals or exceeds one or more thresholds. In some cases, when a neurological condition probability / classification calculation value equals or exceeds an adjustable threshold, the system may display a notification to a subject (e.g., a patient) or user (e.g., a health care practitioner, doctor, nurse, etc.) that the system has detected persistent neurological condition activity. The notification may also include any color. For example, the background of the screen that displays the notification may be red. The text of the notification may be any color, for example, white. The color of the screen background may correlate with the value of the neurological condition probability calculation. For example, if the neurological condition probability is equal to or exceeds a threshold, the selected color for the screen background may indicate that the neurological condition probability is equal to or exceeds the threshold. The color of the notification text may be correlated with the value of the neurological condition probability calculation. For example, if the neurological condition probability calculation is equal to or exceeds a threshold, the selected color for the notification text may indicate that the stroke probability is equal to or exceeds the threshold.

[0202] In some embodiments, the method may include generating one or more notifications upon a positive classification of sedation, delirium, seizure, stroke, or any combination thereof, or upon a probability of sedation, delirium, seizure, stroke, or any combination thereof equaling or exceeding one or more thresholds. In some cases, upon a calculated probability / classification of sedation, delirium, seizure, stroke, or any combination thereof equaling or exceeding an adjustable threshold, the system may display a notification to a subject (e.g., a patient) or user (e.g., a health care practitioner, doctor, nurse, etc.) that the system has detected persistent sedation, delirium, seizure, stroke, or any combination thereof activity.

[0203] The system may also display a wide variety of information to the subject or user in addition to the notification of the detected persistent neurological condition activity. The system may display a neurological condition probability plot 315, the severity of the neurological condition 320, the location of the neurological condition in the subject 335, the likelihood of the neurological condition type occurring, the classification of the neurological condition type 330, the time period over which the persistent neurological condition activity was detected (e.g., 7:40 PM to 7:50 PM), and the like. The one or more notifications may be usable by a health care practitioner to assess the subject's condition and determine a course of treatment. The notification may provide a diagnostic output to the health care practitioner. In some embodiments, the diagnostic output may include a neurological condition classification selected from a plurality of different neurological condition classes or neurological condition types. In some cases, the method may provide one or more neurological condition classifications described elsewhere herein. In some cases, the diagnostic output may provide symptoms related to the neurological condition classification that correlate with a particular set of EEG signals. In some embodiments, the information in the diagnostic output may be usable to identify or detect one or more neurological condition pathologies.

[0204] In some embodiments, one or more notifications (e.g., diagnostic outputs) may be generated when the neurological condition probability value equals or exceeds one or more thresholds, as described elsewhere herein. In some embodiments, one or more notifications (e.g., diagnostic outputs) may be generated when the neurological condition classification of one or more features is completed. In some cases, one or more notifications may be generated in the form of visual, audio, and / or text alerts. The device may include a speaker 325 to provide audio notifications. In some cases, one or more notifications may be delivered via networked communication technologies such as the Internet, telephone, fax, pager, short message service, etc. In some cases, the form, content, or delivery mechanism of the one or more notifications generated may depend on the neurological condition probability value. In some cases, the user may be able to select the form, content, or delivery mechanism of the one or more notifications generated.

[0205] In some embodiments, the neurological condition detection output may include an interface. The interface may provide an indication of EEG signal activity for multiple channels from the data module 110. The interface may display parameters that the user may adjust, such as a time display, a scale, a high pass frequency, a low pass frequency, or a notch value. The interface may also provide a neurological condition probability plot, as described elsewhere herein. The interface may also provide neurological condition probability load results over different time periods. The interface may also depict stroke probability determinations per time segment. The interface may also provide a mechanism for the user to accept or reject the algorithm-derived neurological condition probability determination or neurological condition classification. The interface may also provide a mechanism for the user to input their own determination of a neurological condition, including a segment or a neurological condition classification. In some cases, the neurological condition probability and / or the neurological condition classification may be adjusted as a result of user-entered information regarding a neurological condition episode. The displayed neurological condition probability / neurological condition classification and neurological condition notification may be based solely on algorithm-derived neurological condition decisions, solely on user-entered neurological condition decisions, or a combination of algorithm and user neurological condition decisions.

[0206] In some embodiments, the neurological condition probability / classification calculation module may calculate a neurological condition probability. The neurological condition notification module may output a notification if the neurological condition probability value exceeds a threshold. In some cases, the notification may be generated to a specific person that the method is programmed to notify. The threshold for notification may also be user adjustable.

[0207] In some embodiments, the neurological condition probability / classification calculation and output module may utilize criteria in addition to the neurological condition probability threshold to output a notification. In some cases, dynamic criteria may be applied using a combination of time-based, neurological condition probability-based, and other policies to determine whether a notification is output.

[0208] In some embodiments, the neurological condition probability / classification calculation module may provide a probability value that a neurological condition occurs. The probability value may be provided as a percentage. The probability value may be provided as a scaled value (e.g., 1-10, 0-100, etc.). The probability value may indicate the confidence of the neurological condition and / or the severity of the neurological condition. The probability value may be provided via notification or a time series plot, as described elsewhere herein.

[0209] In some embodiments, the data may further include non-EEG data, which comprises blood pressure, heart rate, or motion data of the subject. In some cases, the EEG data and the non-EEG data are processed in a complementary or synergistic configuration to generate a diagnostic output or improve the accuracy of the diagnostic output. In some cases, the method is performed with data collected from the subject in an external environment outside of a standard healthcare facility. In some cases, the external environment includes a pre-hospital location, a field environment, or an outpatient setting.

[0210] In some embodiments, the system may be coupled with other systems. In some cases, the system may be an eye tracker, a movement sensor (e.g., an accelerometer or inertial measurement unit), an electromyograph (EMG), an electrocardiogram (ECG or EKG), etc. The collection of EEG data may be complemented with other inputs, including other commonly collected biological inputs, such as observed symptoms, cardiac signals (ECG or other heart rate monitors) and blood pressure, and passively collected movement measurements from accelerometers and gyroscopes (for changes in movement, blood flow, and artifact detection), or questionnaire inputs collected from the user. The collection of EEG and complementary input data may be used rapidly in ambulances or other pre-hospital locations, by practitioners in the hospital, for rapid triage in the emergency department (ED), for long-term patient monitoring in the intensive care unit (ICU), and also perioperatively before, during, and after surgery or other in-hospital procedures. For such procedures, separate or continuous recordings may be performed to monitor cerebral deterioration from stroke and other changes in sedation state and patient health, including stroke, any neurological or cardiovascular complications (edema, swelling, etc.), and may allow for more individualized patient-based algorithmic features, processing, tracking, and baseline comparison with baseline patient data (including but not limited to focal slowing, asymmetry, changes in delta and theta-to-delta ratio).

[0211] Complementary non-EEG sensor data can be used for data selection or as independent features in machine learning algorithms. Non-EEG data may be used to complement EEG features and improve the performance of the algorithms.

[0212] In some embodiments, the method may include transmitting the diagnostic output substantially in real time over one or more wired or wireless networks to enable remote stroke management and treatment for the subject.

[0213] In some embodiments, the diagnostic output comprises a single diagnosis, a binary diagnosis, or a multi-hierarchical diagnosis associated with the neurological condition or the onset of one or more neurological conditions.

[0214] In some embodiments, the method may be further scalable and configured for diagnosis of acute traumatic brain injury. In some cases, the EEG signal may include a signal pattern associated with or indicative of asymmetry in different areas and hemispheres of the subject's brain. In some embodiments, the EEG signal is passively collected using a set of electrodes worn on the subject's head. In some cases, the plurality of features includes at least 50 distinct features. In some cases, the plurality of features includes at least 100 distinct features.

[0215] In some embodiments, the system may provide sedation state monitoring across an array of drug agents including, for example, alfentanil, desflurane, fentanyl, isoflurane, nitrous oxide, propofol, remifentanil, sevoflurane, and the like.

[0216] IV. Delirium Detection FIG. 7 diagrammatically illustrates an exemplary delirium detection module 710 intended to analyze previously obtained sections of adult (greater than or equal to 18 years of age) EEG recordings 715 to aid healthcare providers in assessing delirium, according to an embodiment of the present disclosure. The EEG recordings comprise waveforms received from eight channels, each channel carrying an EEG recording from one of eight EEG electrodes (not shown) placed on the subject's head. A pre-processing module 721 is configured to pre-process the incoming EEG recordings and includes a filtering module 723 and a segmentation module 725 to process each of the EEG recordings 715. The eight incoming EEG recordings are band-pass filtered by the filtering module 725 at 1-35 Hz using a 10th order Butterworth filter and divided into temporal segments, each segment having a 10 second duration.

[0217] The preprocessed temporal segments 727 of the EEG recordings are received for further processing by a machine learning module 731. The machine learning module comprises a feature extraction module 733 and a set of machine learning models 735. The feature extraction module 733 is configured to extract a predetermined set of features from each of the preprocessed temporal segments. The feature extraction module 733 may extract a number of different features, including one or more time domain features, one or more frequency domain features, and one or more correlation-based features, as provided above in Tables 1 and 2, for example, or other suitable features. The feature extraction module 733 is configured to extract 59 features from each temporal segment, as shown in FIG. 7, by way of example. Examples of time domain features extracted by the feature extraction module 733 include, but are not limited to, amplitude range, RMS of amplitude, standard deviation of amplitude, sharpness, area under the wave, number of minima and / or maxima, peak amplitude, zero crossings, RMS of derivative of the signal, regularity. Examples of frequency domain features extracted by the feature extraction module 733 include, but are not limited to, dominant frequency, dominant frequency power, signal leakage outside the dominant frequency, spectral entropy, power of the signal within a given frequency band (e.g., alpha band, beta band, gamma band, delta band, or theta band). Examples of correlation-based features optionally include correlation of a channel relative to other channels on the same hemisphere of the brain.

[0218] Each of the machine learning models 735 is a random forest model configured to classify a temporal segment as delirium positive or delirium negative based on a set of features extracted from the temporal segment. Each random forest model is an ensemble of 30 bagged decision trees that are separately generated to classify a temporal segment of the EEG recordings from one of the channels.

[0219] Each temporal segment may also be evaluated by an artifact removal module 737. Some combination of features extracted by the feature extraction module 733 for each temporal segment is used by the artifact removal module 737 to determine whether a given temporal segment comprises an artifact signal or excessive artifacts in the signal and should be excluded from further analysis to contribute to the final delirium assessment. A given EEG electrode typically operates at a certain characteristic impedance or within an impedance range when the electrode is working normally, and impedance measurements that deviate from that impedance or impedance range indicate a defective or damaged electrode. The artifact removal module 737 may also use the impedance measurements of each electrode to determine whether a temporal segment should be excluded from further analysis.

[0220] Each individual output of the random forest model 735 within a time window, which may encompass one or more consecutive time epochs, is combined and processed against a set of predefined rules, shown diagrammatically as a control policy 741, to determine the subject's overall delirium score. The overall delirium score may be transmitted to other modules, devices, or components of the system as an output of the control policy 743. As an example, the overall delirium score may be a value based on the percentage of time segments within the time window that are delirium positive, such that a higher percentage of time segments that are classified as delirium positive results in a higher delirium burden. The time window may be a moving time window, such that the overall delirium score output 743 is a trace or plot of the overall delirium score over time and / or is dynamically updated. Alternatively or additionally, the output 743 may comprise an alert that is generated when the overall delirium score exceeds a predetermined threshold.

[0221] When the module 710 was used in a pilot study, the algorithm employed for delirium detection achieved clinical success in detecting delirium, resulting in a sensitivity of about 95% and a specificity of about 92%, as further described in Example 1. In addition, continuous monitoring for delirium was achieved as soon as the electrodes began recording the EEG waveform. By providing continuous monitoring for delirium, the module 710 may optimize delirium treatment and thus reduce the duration of delirium.

[0222] FIG. 8 diagrammatically illustrates an alternative exemplary delirium detection module 810 intended to analyze previously obtained sections of adult (greater than or equal to 18 years of age) EEG recordings 815 to aid healthcare providers in assessing delirium, according to an embodiment of the present disclosure. The EEG recordings comprise waveforms received from eight channels, each channel carrying an EEG recording from one of eight EEG electrodes (not shown) placed on the subject's head. A pre-processing module 821 is configured to pre-process the incoming EEG recordings and includes a filtering module 823 and a segmentation module 825 to process each of the EEG recordings 815. The eight incoming EEG recordings are band-pass filtered at 0.5-40 Hz using a 5th order Butterworth filter by the filtering module 823 and then segmented into temporal segments by the segmentation module 825. Each temporal segment has a duration of 60 seconds. In addition, the first half (30 seconds) of the time segment overlaps with the previous time segment, and the second half of the time segment overlaps with the subsequent time segment.

[0223] The preprocessed temporal segments 827 of the EEG recording are received for further processing by a machine learning module 831. The machine learning module comprises a single channel feature extraction module 833 and a multi-channel feature extraction module 834. The single channel feature extraction module 833 may be configured to extract a predetermined set of features from each of the preprocessed temporal segments 827. The single channel feature extraction module 833 may extract a number of different features, including one or more time domain features and one or more frequency domain features. Examples of time domain features extracted by the feature extraction module 833 include, but are not limited to, amplitude range, RMS of amplitude, standard deviation of amplitude, sharpness, area under the wave, number of minima and / or maxima, peak amplitude, zero crossings, RMS of derivative of the signal, regularity. Examples of frequency domain features extracted by the feature extraction module 833 include, but are not limited to, dominant frequency, dominant frequency power, leakage of signals outside the dominant frequency, spectral entropy, power of signals within a given frequency band (e.g., alpha band, beta band, gamma band, delta band, or theta band).

[0224] The multi-channel feature extraction module 834 is configured to extract a predetermined set of multi-channel features that quantify the degree of correlation between paired temporal segments from different EEG signals corresponding to a given time epoch. Unlike single-channel features that characterize a given temporal segment from an EEG signal received from one channel, multi-channel features characterize the inter-channel interactions within a given time epoch. Examples of multi-channel features include, but are not limited to, the average correlation coefficient for EEG signal waveforms received from paired electrodes within and / or across a hemisphere (the waveforms may be filtered to isolate signals within alpha, beta, gamma, delta, or theta frequency bands), the average peak correlation within and / or across a hemisphere (measured over different lags), the average lags at which peak correlations are observed within and / or across a hemisphere, and the average correlation coefficient of power spectra within and / or across a hemisphere.

[0225] The machine learning module 831 further comprises a set of single channel machine learning models 835 configured to receive and process the single channel features and provide delirium probability per time segment, and a multi-channel machine learning model 836 configured to process the multi-channel features and provide delirium probability per time epoch. Each single channel machine learning model 835 is a random forest model that is an ensemble of 30 binary decision trees (classifying EEG windows as delirium positive or negative) generated separately for one of the channels to evaluate the time segment of the EEG recording. Each model is trained on an individual channel using random undersampling to predict delirium probability over a given time segment.

[0226] The multi-channel machine learning models 836 are boosted random forest models that are trained on the multi-channel features to predict delirium probability over a given time epoch. Each boosted random forest model comprises an ensemble of 50 binary decision trees that classify EEG windows as delirium positive or negative.

[0227] Each temporal segment may also be evaluated by an artifact removal module 837. For each temporal segment, some combination of features extracted by the single channel feature extraction module 833 is used by the artifact removal module 837 to determine whether a given temporal segment comprises an artifact signal or excessive artifacts in the signal and should be excluded from further analysis to contribute to the final delirium assessment. The artifact removal module 837 may also use the impedance measurements of each electrode to determine whether a temporal segment should be excluded from further analysis.

[0228] The individual outputs of each of the single channel machine learning model 835 and the multi-channel machine learning model 836 within a time window are combined and processed against a set of predefined rules, shown diagrammatically as control policy 841, to determine the subject's overall delirium score. The time window may encompass one or more consecutive time epochs.

[0229] The overall delirium score, as determined by the control policy 841, may be transmitted to other modules, devices, or components of the system. As an example, the overall delirium score may be a value that is a weighted combination of the delirium probabilities determined by the single channel machine learning model 835 and the multi-channel machine learning model 836.

[0230] The time window may be a moving time window such that the overall delirium score may be presented as a trace or plot of the overall delirium score over time and / or dynamically updated. Alternatively, or in addition, the control policy 841 may be configured to generate an alert when the overall delirium score exceeds a predetermined threshold.

[0231] In an embodiment, the delirium detection module 710 or the delirium detection module 810 may be included within the processing module 120, as shown in FIGURE 2 and described herein above. In an embodiment, the delirium detection module 710 or the delirium detection module 810 may operate within the processing module 120 along with other detection modules configured to assess one or more other neurological conditions, such as, but not limited to, sedation, stroke, or seizures.

[0232] FIG. 9 diagrammatically illustrates another exemplary delirium detection module 900 for analyzing previously obtained sections of EEG recordings 915 to aid a healthcare provider in assessing delirium using a machine learning model 931. The EEG recordings may be obtained from multiple electrodes, which may be coupled to or incorporated into a headband, headgear, or other device configured to place the electrodes on or around the patient's head. In some variations, the EEG recordings are obtained from 10 electrodes coupled to or incorporated into the headband. In other variations, the EEG recordings are obtained from 16 electrodes coupled to or incorporated into the headband. Similar to module 810 in FIG. 8, the delirium detection module 900 may include a pre-processing module 921 configured to pre-process the incoming EEG recordings. The pre-processing module 921 may include a filtering module 923 and a segmentation module 925 to process each of the EEG recordings 915. The incoming EEG recording may be band-pass filtered at 0.5-40 Hz using a 5th order Butterworth filter by filtering module 923 and then segmented into temporal segments by segmentation module 925, generating pre-processed temporal segments 927. Each temporal segment may have a duration of 60 seconds.

[0233] The pre-processed temporal segments 927 of the EEG recording are received for further processing by a machine learning module 931. The machine learning module 931 may comprise a single channel feature extraction module 903 and a multi-channel feature extraction module 905. The single channel feature extraction module 903 may be configured to extract a predetermined set of features from each of the pre-processed temporal segments 927. The single channel feature extraction module 903 may extract a number of different features, including one or more time domain features and one or more frequency domain features. Examples of time domain features extracted by the feature extraction module 903 include, but are not limited to, amplitude range, RMS of amplitude, standard deviation of amplitude, sharpness, area under the wave, number of minima and / or maxima, peak amplitude, zero crossings, RMS of derivative of the signal, regularity. Examples of frequency domain features extracted by the feature extraction module 903 include dominant frequency, dominant frequency power, leakage of signals outside the dominant frequency, spectral entropy, power of signals within a given frequency band (e.g., alpha, beta, gamma, delta, or theta band). Different features extracted may also include power within different frequency bands (e.g., alpha, beta, delta, theta, and gamma), spectral properties, power ratios, amplitude characteristics and morphological structure features, entropy, variability, and wavelet decomposition.

[0234] The multi-channel feature extraction module 905 may be configured to extract a predetermined set of multi-channel features that quantify the degree of correlation between paired temporal segments from different EEG signals corresponding to a given time epoch. Unlike single-channel features that characterize a given temporal segment from an EEG signal received from one channel, multi-channel features characterize the inter-channel interactions within a given time epoch. Examples of multi-channel features include, but are not limited to, average correlation coefficients for EEG signal waveforms received from paired electrodes within and / or across hemispheres (waveforms may be filtered to isolate signals within alpha, beta, gamma, delta, or theta frequency bands), average peak correlations within and / or across hemispheres (measured over different lags), average lags at which peak correlations are observed within and / or across hemispheres, and average correlation coefficients of power spectra within and / or across hemispheres. In some variations, multi-channel features that may be calculated to quantify inter-channel interactions may include intra- and / or cross-hemispheric correlations for different frequency bands (e.g., alpha, beta, delta, theta, and gamma), as well as spectral, amplitude and phase related correlations, and synchrony measures. The multi-channel machine learning model 905 may be a boosted random forest model trained on the multi-channel features to predict delirium probability over a given time epoch. Each boosted random forest model may comprise an ensemble of 50 binary decision trees that classify EEG windows as delirium positive or negative.

[0235] In model 931 of FIG. 9, 16 (identical to the number of EEG channels) boosted random forests 902 may be trained on individual channel features using random undersampling to predict delirium probability for every 60 second EEG window. Simultaneously, boosted random forests 904 may be trained on multi-channel features to predict delirium probability. Each boosted random forest may consist of an ensemble of 50 binary decision trees (classifying EEG windows as delirium positive or negative). Weighted combinations of probability measures obtained from the random forest models may be non-linearly combined to obtain a delirium score between 0 (indicating low delirium probability) and 1 (indicating high delirium probability). An exponentially weighted smoothing filter may be applied to the delirium probability score to produce a single value (also between 0 and 1), which can be used as a diagnostic tool for delirium.

[0236] Boosting can increase emphasis when the model is having difficulty learning, allowing for better overall predictive ability for the model. At the same time, random undersampling of the data (e.g., 80% of the minority class available for training) can allow for the selection of a subset of the data such that the number of delirium positives and negatives visible to the model for training can be equalized.

[0237] Each temporal segment may also be evaluated by an artifact removal module 937. For each temporal segment, some combination of features extracted by the single channel feature extraction module 903 may be used by the artifact removal module 937 to determine whether a given temporal segment comprises an artifact signal or excessive artifacts in the signal and should be excluded from further analysis to contribute to the final delirium assessment. The artifact removal module 937 may also use the impedance measurements of each electrode to determine whether a temporal segment should be excluded from further analysis.

[0238] The individual outputs of the single channel machine learning model 903 and the multi-channel machine learning model 905 within a time window may be combined and processed against a set of predefined rules, shown diagrammatically as a control policy 906, to determine an overall delirium score for the subject. The time window may encompass one or more consecutive time epochs.

[0239] The overall delirium score, as determined by the control policy 906, may be transmitted to other modules, devices, or components of the system. As an example, the overall delirium score may be a value that is a weighted combination of the delirium probabilities determined by the single channel machine learning model 903 and the multi-channel machine learning model 905.

[0240] V. After neurological condition detection In some embodiments, the method may include generating one or more notifications as described elsewhere herein.

[0241] In some embodiments, the method may provide a user with a response to minimize or prevent the detected neurological condition. The method may provide a response to minimize or reduce the risk of developing a neurological condition. In some cases, a therapy may be delivered to the subject to prevent and / or alleviate the predicted neurological condition. In some cases, the method may adjust the amount of therapy delivered to the subject.

[0242] VI. Computer Systems The present disclosure provides a computer system that is programmed to implement the methods of the present disclosure, including controlling the multi-detection system, controlling hardware components, receiving and processing data, interfacing with a user, etc. FIG. 6 shows a computer system 601 that is programmed or otherwise configured to operate and / or control the data and processing modules. The computer system 601 can coordinate various aspects of the present disclosure, such as determining a neurological condition, classifying a neurological condition, classifying EEG signals, classifying seizures, classifying delirium, classifying strokes, classifying sedation, generating notifications, generating probability plots of neurological conditions, processing EEG signals, segmenting EEG signals, extracting features, processing features using machine learning algorithms, associating features with assessment scales, implementing control policies and neurological condition loads, calculating values, plotting probabilities of neurological conditions, etc. The computer system 601 can be a user's electronic device or a computer system located remotely relative to the electronic device. The electronic device can be a mobile electronic device.

[0243] The computer system 601 includes a central processing unit (CPU, also referred to herein as "processor" and "computer processor") 605, which can be a single-core or multi-core processor or multiple processors for parallel processing. The computer system 601 also includes a memory or memory location 610 (e.g., random access memory, read-only memory, flash memory), an electronic storage unit 615 (e.g., hard disk), a communication interface 620 (e.g., network adapter) for communicating with one or more other systems, and peripheral devices 625, such as cache, other memory, data storage devices, and / or electronic display adapters. The memory 610, the storage unit 615, the interface 620, and the peripheral devices 625 communicate with the CPU 605 through a communication bus (solid lines) such as a motherboard. The storage unit 615 can be a data storage unit (or data repository) for storing data. The computer system 601 can be operatively coupled to a computer network ("network") 630 with the aid of the communication interface 620. The communication interface may be wired or wireless. The network 630 may be the Internet, the Internet and / or an extranet, or an intranet and / or an extranet in communication with the Internet. The network 630 may in some cases be a telecommunications and / or data network. The network 630 may include one or more computer servers, which may enable distributed computing such as cloud computing. The network 630 may in some cases implement a peer-to-peer network with the help of the computer system 601, which may enable devices coupled to the computer system 601 to behave as clients or servers.

[0244] CPU 605 can execute sequences of machine-readable instructions, which may be embodied in a program or software. The instructions may be stored in a memory location, such as memory 610. The instructions can be directed to CPU 605, which can then be programmed or otherwise configured to implement the methods of the present disclosure. Examples of operations performed by CPU 605 can include fetching, decoding, executing, and writeback.

[0245] The CPU 605 can be part of a circuit, such as an integrated circuit. One or more other components of the system 601 can be included in the circuit. In some cases, the circuit is an application specific integrated circuit (ASIC).

[0246] The storage unit 615 can store files such as drivers, libraries, and saved programs. The storage unit 615 can store user data, such as user preferences and user programs. The computer system 601 can include one or more additional data storage units that are, in some cases, external to the computer system 601, such as located on a remote server that communicates with the computer system 601 through an intranet or the Internet.

[0247] Computer system 601 can communicate with one or more remote computer systems through network 630. For example, computer system 601 can communicate with a user's remote computer system (e.g., a neurological condition detection system management device, a neurological condition detection system user, a neurological condition data acquisition device, a neurological condition detection system recording device). Examples of remote computer systems include a personal computer (e.g., a portable PC), a slate or tablet PC (e.g., an Apple® iPad®, a Samsung® Galaxy Tab), a phone, a smartphone (e.g., an Apple® iPhone®, an Android®-enabled device, a BlackBerry®), or a personal digital assistant. A user can access computer system 601 via network 630.

[0248] Methods as described herein may be implemented using machine (e.g., computer processor) executable code stored on electronic storage locations of computer system 601, such as, for example, on memory 610 or electronic storage unit 615. The machine executable or machine readable code may be provided in the form of software. During use, the code may be executed by processor 605. In some cases, the code may be read from storage unit 615 and stored on memory 610 for easy access by processor 605. In some circumstances, electronic storage unit 615 may be omitted and machine executable instructions are stored on memory 610.

[0249] The code can be pre-compiled and configured for use with a machine having a processor adapted to execute the code, or can be compiled during run-time. The code can be provided in a programming language that can be selected to allow the code to be executed in a pre-compiled or as-compiled manner.

[0250] Aspects of the systems and methods provided herein, such as the computer system 601, can be embodied in programming. Various aspects of the technology may be considered as "products" or "articles of manufacture" in the form of machine (or processor) executable code and / or associated data, typically carried on or embodied in some type of machine-readable medium. The machine executable code may be stored on an electronic storage unit such as memory (e.g., read-only memory, random access memory, flash memory) or hard disk. A "storage" type medium may include any tangible memory of a computer, processor, or the like, or its associated modules, such as various semiconductor memories, tape drives, disk drives, and the like, which may provide non-transitory storage for software programming from time to time. All or portions of the software may be communicated from time to time over the Internet or various other telecommunications networks. Such communication may, for example, enable loading of the software from one computer or processor to another, for example, from a management server or host computer to a computer platform of an application server. Thus, other types of media that may carry software elements include optical, electrical, and electromagnetic waves, such as those used across physical interfaces between local devices, through wired and optical terrestrial networks, and via various wireless links. Physical elements that carry such waves, such as wired or wireless links, optical links, or the like, may also be considered software-bearing media. As used herein, unless limited to non-transitory tangible "storage" media, terms such as computer or machine "readable medium" refer to any medium that participates in providing instructions to a processor for execution.

[0251] Thus, a machine-readable medium such as a computer executable code may take many forms, including, but not limited to, a tangible storage medium, a carrier wave medium, or a physical transmission medium. Non-volatile storage media include optical or magnetic disks, such as any of the storage devices in any computer or equivalent, such as may be used to implement the databases, etc., shown in the figures. Volatile storage media include dynamic memory, such as the main memory of such a computer platform. Tangible transmission media include coaxial cables, copper wire, and optical fibers, including the wires that comprise a bus within a computer system. Carrier wave transmission media may take the form of electric or electromagnetic signals or acoustic or light waves, such as those generated during radio frequency (RF) and infrared (IR) data communications. Common forms of computer readable media thus include, for example, a floppy disk, a flexible disk, a hard disk, a magnetic tape, any other magnetic medium, a CD-ROM, a DVD or DVD-ROM, any other optical medium, punch cards paper tape, any other physical storage medium with a pattern of holes, RAM, ROM, PROM and EPROM, Flash EPROM, any other memory chip or cartridge, a carrier wave carrying data or instructions, a cable or link carrying such a carrier wave, or any other medium from which a computer may read programming code and / or data. Many of these forms of computer readable media may be involved in carrying one or more sequences of one or more instructions to a processor for execution.

[0252] The computer system 601 can include or communicate with an electronic display 635 with a user interface (UI) 640, for example, programmed to control the multi-indication detection system and functionality, to provide a login screen for administrator access to the software, and / or to provide operational status health of the multi-indication detection system. Examples of UIs include, but are not limited to, graphical user interfaces (GUIs) and web-based user interfaces.

[0253] The methods and systems of the present disclosure can be implemented using one or more algorithms. The algorithms can be implemented using software, responsive to execution by the central processing unit 605. The algorithms can be, for example, software components described elsewhere herein, and may modulate seizure detection system parameters (e.g., EEG signal processing, machine learning algorithms, control strategies, neurological condition loading, notifications, etc.).

[0254] Working Example Clinical studies using delirium detection monitors A pilot clinical study was conducted using the delirium detection monitor illustrated in FIG. 7. The study obtained EEG recordings from a total of 94 subjects. The first cohort of subjects consisted of 81 ICU patients undergoing EEG recordings with the delirium detection monitor as part of their routine clinical procedure. Of the 81 ICU patients, 69 were delirium positive and 12 were delirium negative. For these patients, delirium assessment was performed by a trained clinical nurse with experience in delirium assessment using the CAM-ICU. The second cohort of subjects consisted of 13 healthy volunteers undergoing EEG recordings with the delirium detection monitor in a research setting that simulated hospital conditions. All 13 healthy volunteer subjects were delirium negative. As further detailed in Table 1 below, the 94-subject pilot study dataset was split into two groups: approximately 2 / 3 of the subjects (63 subjects) were used for delirium detection monitor algorithm development and training, and approximately 1 / 3 of the subjects (31 subjects) were used for algorithm validation. The datasets were strictly separated; there was no crossover of subjects between the development and validation datasets, and data from the validation dataset was not used for algorithm training. [Table 1]

[0255] The performance of the delirium detection monitor was evaluated by calculating the sensitivity and specificity for correct identification of delirium positive patients. The performance from the validation dataset is shown below in Table 2. As shown in the table, the use of the delirium detection monitor achieved clinical success, resulting in a sensitivity of approximately 95% and a specificity of approximately 92% in detecting delirium. [Table 2]

[0256] The foregoing description uses specific terminology for purposes of explanation to provide a thorough understanding of the invention. However, it will be apparent to one skilled in the art that specific details are not required to practice the invention. Thus, the foregoing descriptions of specific embodiments of the invention are presented for purposes of illustration and description. They are not intended to be exhaustive or to limit the invention to the precise forms disclosed. Obviously, many modifications and variations are possible in light of the above teachings. The embodiments have been chosen and described in order to explain the principles of the invention and its practical application, thereby enabling those skilled in the art to utilize the invention and its various embodiments with various modifications as may be suitable for the particular use contemplated.

Claims

1. 1. A method for detecting delirium, the method being executed on a computer system comprising one or more processors, The method comprises: The one or more processors acquire data from a subject, the data comprising a plurality of electroencephalography (EEG) signals recorded across a plurality of channels; The one or more processors pre-process the data, wherein pre-processing the data includes: Dividing the EEG signal into a plurality of temporal segments, each temporal segment corresponding to a time epoch defined by at least a start time and a duration; extracting a plurality of features from each of the plurality of temporal segments; and generating, by the one or more processors, a delirium classification for each of the temporal segments based on the extracted features using one or more machine learning models; the one or more processors determining an overall delirium score for the subject for a time window, the overall delirium score being based on the delirium classification generated by the one or more machine learning models, the time window comprising one or more time epochs; A method comprising:

2. 2. The method of claim 1, wherein the delirium classification is a binary classification of delirium positive or delirium negative, a delirium probability value, or a delirium severity value.

3. The method of claim 1, further comprising the one or more processors providing a trace of the overall delirium score over time.

4. The method of claim 3, further comprising the one or more processors determining a trend line for the trace.

5. 10. The method of claim 1, wherein preprocessing the data further comprises extracting a plurality of multi-channel features from different EEG signals corresponding to a given time epoch that quantify the degree of correlation between paired time segments.

6. The method of claim 5, further comprising the one or more processors using a multi-channel machine learning model to generate a multi-channel delirium classification for each time epoch based on the plurality of multi-channel features, and the delirium score is further based on the multi-channel delirium classification.

7. 10. The method of claim 1, wherein the delirium is hypoactive delirium.

8. The method of claim 1 , wherein the time window has a duration that encompasses one time epoch.

9. The method of claim 1 , wherein the time window has a duration that encompasses a plurality of consecutive time epochs.

10. The method of claim 1 , wherein the duration of each of the time epochs ranges from about 1 second to about 10 minutes.

11. 11. The method of claim 10, wherein the duration of each of the time epochs is about 10 seconds, about 30 seconds, about 60 seconds, about 2 minutes, about 5 minutes, or about 10 minutes.

12. The method of claim 1, wherein successive time epochs do not overlap.

13. The method of claim 1, wherein successive time epochs overlap by less than 50%.

14. The method of claim 1 , wherein the plurality of features comprises at least one time-domain feature.

15. The method of claim 1 , wherein the plurality of features comprises at least one frequency domain feature.

16. 10. The method of claim 1, wherein the plurality of features comprises at least one feature quantifying a degree of correlation between at least one of the plurality of temporal segments and a corresponding time-based segment of at least one other simultaneously collected EEG signal.

17. 17. The method of claim 16, wherein the EEG signal from the at least one of the plurality of time segments and the at least one other simultaneously collected EEG signal are collected from the same hemisphere of the brain.

18. The method of claim 1 , wherein each channel is assigned to an independent machine learning model, and for each channel, the extracted features are applied to the machine learning model corresponding to the channel.

19. The method of claim 1 , wherein the one or more machine learning models are random forest models.

20. The method of claim 1 , wherein the plurality of EEG signals are obtained from a plurality of electrodes incorporated into a headband.

21. The method of claim 1 , wherein the plurality of channels comprises eight channels.

22. The method of claim 1 , wherein the plurality of channels comprises 16 channels.

23. The method of claim 1 , wherein the detected delirium is hypoactive delirium.

24. 1. A system for detecting delirium, comprising: a data module configured to receive data from a subject, the data comprising a plurality of electroencephalography (EEG) signals recorded across a plurality of channels during a time window; a delirium detection module comprising a memory storing a set of instructions and one or more processors; Equipped with The one or more processors, in response to the set of instructions, Pre-processing the data received by the data module, wherein pre-processing the data includes: Dividing the EEG signal into a plurality of temporal segments, each temporal segment corresponding to a time epoch defined by at least a start time and a duration; extracting a plurality of features from each of the plurality of temporal segments; and generating a delirium classification for each of the temporal segments based on the extracted features using one or more machine learning models; determining an overall delirium score based on the delirium classification generated by the one or more machine learning models; and A system configured to:

25. 25. The system of claim 24, wherein the plurality of channels comprises eight channels.

26. 25. The system of claim 24, wherein the plurality of channels comprises 16 channels.

27. 25. The system of claim 24, wherein each channel of the plurality of channels is assigned to an independent machine learning model, and for each channel, the extracted features are applied to the machine learning model corresponding to each channel.

28. 25. The system of claim 24, wherein the one or more machine learning models comprise a random forest model.

29. 25. The system of claim 24, further comprising a headband, the headband comprising a plurality of electrodes, and the plurality of electroencephalography (EEG) signals being recorded from the plurality of electrodes.

Images