阅读说明：本技术 用于针对生物时间序列利用经验零假设的方法和系统 (Method and system for utilizing empirical zero assumptions for biological time series ) 是由 S·B·威尔森 M·朔尔 于 2019-03-11 设计创作，主要内容包括：本文公开了一种用于针对生物时间序列利用经验零假设的方法和系统。方法(100)还包括使用大规模检验来估计针对时期的子集的经验零假设,从而计算第二多个时期与第一多个时期相似的量。方法(100)还包括确定第二多个时期是否来自相同的EEG记录。(Disclosed herein is a method and system for utilizing an empirical zero hypothesis for biological time series. The method (100) further includes estimating empirical zero hypotheses for a subset of epochs using a large-scale test to calculate an amount by which the second plurality of epochs are similar to the first plurality of epochs. The method (100) further includes determining whether the second plurality of epochs are from the same EEG recording.)

1. A method for determining whether a first portion of a biological time series is comparable to a second portion of the biological time series, the method comprising:

selecting a first portion of a biological time series;

establishing the first portion as a baseline portion;

selecting a second portion for measurement; and

an empirical zero assumption is used to determine the similarity of the second portion to the baseline portion.

2. A method for determining whether two sets of epochs are comparable, the method comprising:

selecting a first plurality of epochs of the EEG recording as baseline segments;

generating a second plurality of epochs;

estimating empirical zero hypotheses for a subset of the epochs using a large-scale test to calculate quantities that are similar for the second plurality of epochs to the first plurality of epochs; and

determining whether the second plurality of epochs are from the EEG recording.

3. An EEG system, the EEG system comprising:

a processor communicatively coupled to a computer-readable medium, wherein the computer-readable medium comprises instructions executable by the processor to:

selecting a first epoch of an EEG recording;

establishing the first time period as a baseline time period;

Selecting a second time period for measurement; and

an empirical zero assumption is used to determine the similarity of the second epoch to the baseline epoch.

Technical Field

The present invention generally relates to a method and system for utilizing an empirical zero hypothesis for biological time series.

Background

Electroencephalography ("EEG") is a diagnostic tool that measures and records the electrical activity of the human brain in order to assess brain function. A plurality of electrodes are attached to the head of the person and connected to the machine by wires. The machine amplifies the signal and records the electrical activity of the human brain. Electrical activity is generated by the sum of neural activity across multiple neurons. These neurons generate small voltage fields. The summation of these voltage fields creates an electrical reading that can be detected and recorded by the electrodes on the person's head. EEG is a superposition of multiple simpler signals. In normal adults, the amplitude of the EEG signal is typically in the range of 1 to 100 microvolts, and the EEG signal is about 10 to 20 millivolts when measured with subdural electrodes. The detection of the amplitude and temporal dynamics of the electrical signal provides information about the underlying (underlying) neural activity and medical condition of the person.

EEG is performed for: diagnosing epilepsy; problems with dementia or loss of consciousness are identified; verifying brain activity of a person in coma; study sleep disorders, monitor brain activity during surgery, and other physical problems.

During EEG a number of electrodes (typically 17-21, but there are at least 70 standard positions) are attached to a person's head. The electrodes are referenced by their position relative to a lobe (lobe) or region of the human brain. Reference is made to the following: f is frontal lobe; fp is frontal pole; t-temporal lobe; c is medium leaf; p is apical leaf; o ═ occipital leaf; a ═ auricle (ear electrode). The numbers are used to further narrow the position and the "z" point is related to the electrode position in the midline of the person's head. An electrocardiogram ("EKG") may also appear on the EEG display screen.

EEG uses a combination of various electrodes called montages to record brain waves from different amplifiers. Montages are often created to provide a clear picture of the spatial distribution of EEG across the cerebral cortex. A montage is an electrogram obtained from a spatial array of recording electrodes and preferably refers to a particular combination of electrodes examined at a particular point in time.

In a bipolar montage, successive pairs of electrodes are linked by connecting electrode input 2 of one channel to input 1 of a subsequent channel, so that adjacent channels have a common one of the electrodes. The bipolar chains of electrodes may be connected from front to back (longitudinal) or from left to right (lateral). In a bipolar montage, the signals between two active electrode positions are compared, resulting in a difference in activity being recorded.

Another type of montage is the reference montage or monopolar montage. In the reference montage, the various electrodes are connected to input 1 of each amplifier, and the reference electrode is connected to input 2 of each amplifier. In the reference montage, a signal is collected at the active electrode location and compared to a common reference electrode.

The reference montage is useful for determining the true amplitude and morphology of the waveform. For the temporal lobe electrode, CZ is usually a good scalp reference.

Being able to locate the origin of the electrical activity ("location") is critical to being able to analyze the EEG. The localization of normal or abnormal brain waves in a bipolar montage is typically accomplished by identifying "phase reversals," the deflections of two channels in the chain that point in opposite directions. In the reference montage, all channels may show a deflection in the same direction. If the electrical activity at the active electrode is positive when compared to the activity at the reference electrode, the deflection will be downward. An electrode in which the electrical activity is the same as the activity at the reference electrode will not show any deflection. In general, the electrode with the largest upward deflection represents the largest negative activity in the reference montage.

Some patterns indicate a tendency of a person to have seizures. These waves may be referred to by the physician as "epileptiform abnormalities" or "epileptic waves". These include spikes, and both spike and wave discharges. Spikes and spikes in a particular region of the brain (such as the left temporal lobe) indicate that partial seizures may have originated in that region. On the other hand, primary generalized epilepsy is caused by spikes and wavy discharges widely distributed over both hemispheres of the brain, especially if they start simultaneously in both hemispheres.

There are several types of brainwaves: alpha waves, beta waves, theta waves, and gamma waves. The alpha wave has a frequency of 8 to 12 hertz ("Hz"). Alpha waves are typically found when a person is relaxed or when the person is awake with closed eyes but mentally alert.

The alpha wave stops when the eyes of the person are open or when the person is concentrating. The beta wave has a frequency of 13Hz to 30 Hz. Beta waves are commonly found when a person is alert, thinking, excited, or has taken high doses of certain drugs. The wave has a frequency of less than 3 Hz. Waves are usually found only when a person is asleep (non-REM or dreaminess sleep) or the person is a young child. The theta wave has a frequency of 4Hz to 7 Hz. Theta waves are usually found only when a person is asleep (dreaming or REM sleep) or the person is a young child. The gamma wave has a frequency of 30Hz to 100 Hz. Gamma waves are commonly found during higher mental activities and motor functions.

The following sets forth general definitions of terms used in the relevant art.

Boolean algebra is a sub-domain of algebra in which the values of variables are true values true and false, generally denoted 1 and 0, respectively.

Boolean Networks (BN) are mathematical models of biological systems based on boolean logic. BN has a network structure consisting of nodes corresponding to genes or proteins. Each node in BN takes the value 1 or 0, meaning that the gene is expressed or not expressed.

Fuzzy logic is a form of multi-valued logic; it deals with not fixed and precise reasoning but rather approximate reasoning. Fuzzy logic variables may have truth values to the extent that they range between 0 and 1, as compared to traditional binary sets (where the variables may take either true or false values). Fuzzy logic has been extended to handle the concept of partial truth values, where truth values may range between completely true and completely false. Further, when linguistic variables are used, these degrees may be managed by specific functions. Irrational can be described by the so-called "fuzziness".

Multi-layer perceptrons ("MLPs") are feed-forward artificial neural network models that map an input data set onto an appropriate output set. MLP consists of multiple layers of nodes in a directed graph, each layer being fully connected to the next layer. Each node, except for the input nodes, is a neuron (or processing element) with a nonlinear activation function.

Neural networks ("NNs") are interconnected groups of natural or artificial neurons that use mathematical or computational models to process information based on a connection-oriented computational approach. In more practical terms, neural networks are non-linear statistical data modeling or decision tools. They can be used to model complex relationships between inputs and outputs or to find patterns in data.

Perceptrons are simple models of artificial neurons that can predict boolean events after training on past events. The perceptron is specified by the number of inputs N and the weight connecting the inputs to the output nodes. Weights are parameters that must be set manually or learned through a learning algorithm.

The ROC curve (receiver operating characteristic) is a graphical plot with test sensitivity as the y coordinate and 1 minus the specificity or False Positive Rate (FPR) as the X coordinate. The ROC curve is an effective method of assessing the performance of diagnostic tests.

"amplitude" refers to the measured vertical distance from the trough to the maximum peak (negative or positive). It expresses information about the size of the neuron population (population) and its activation synchronicity during the generation of the components.

The term "analog-to-digital conversion" refers to when an analog signal is converted to a digital signal, which can then be stored in a computer for further processing. The analog signal is a "real world" signal (e.g., a physiological signal such as an electroencephalogram, electrocardiogram, or electrooculogram). In order for them to be stored and manipulated by a computer, these signals must be converted into discrete digital form that the computer can understand.

"artifacts" are electrical signals detected by the EEG along the scalp, but they originate from non-brain origins. There are patient-related artifacts (e.g., movement, sweating, ECG, eye movement) and technical artifacts (50/60Hz artifact, cable movement, electrode paste related).

The term "differential amplifier" refers to the key to the electrophysiological equipment. It amplifies the difference between the two inputs (one amplifier per pair of electrodes).

The "duration" is the time interval from the beginning of the voltage change to its return to baseline. It is also a measure of the synchronous activation of neurons involved in component generation.

"electrode" refers to a conductor used to establish electrical contact with a non-metallic portion of an electrical circuit. EEG electrodes are small metal discs, typically made of stainless steel, tin, gold or silver covered with a silver chloride coating. They are placed in specific positions on the scalp.

The "electrode gel" acts as a malleable (mallable) extension of the electrodes so that movement of the electrode leads is less likely to create artifacts. The gel can be maximally in contact with the skin and allows low resistance recordings through the skin.

The term "electrode positioning" (10/20 system) refers to the standardized placement of scalp electrodes for classical EEG recordings. The nature of the system is a 10/20 range of percent distance between the Nasion-Inion and the fixation point. These points are labeled as frontal pole (Fp), middle lobe (C), parietal lobe (P), occipital lobe (O), and temporal lobe (T). The centerline electrode is labeled with a subscript z, which represents zero. Odd numbers are used as subscripts for points on the left hemisphere, while even numbers are used as subscripts for points on the right hemisphere.

"electroencephalography" or "EEG" refers to the trace of brain waves drawn by electroencephalography by recording the electrical activity of the brain from the scalp.

"electroencephalograph" refers to a device for detecting and recording brain waves (also known as a encephalography machine).

"epileptiform" refers to the condition analogous to epilepsy < epileptiform abnormality >.

"filtering" refers to the process of removing unwanted frequencies from a signal.

A "filter" is a device that alters the frequency content of a signal.

An "ideal frequency selective filter" is a filter that passes signals at exactly one set of frequencies, while rejecting the rest completely. There are three types of filters: "Low frequency" or the old term "high pass," filtering low frequencies; high frequency or the old term low pass. Filtering high frequency; a "notch filter" filters a frequency, typically 60 Hz. A "real filter" or "hardware filter" will alter the frequency content of the signal. After filtering the signal, the frequencies that have been filtered will not be recoverable. The "digital filter" changes the frequency of the signal by performing calculations on the data.

"frequency" refers to rhythmic repetitive activity (in Hz). The frequency of EEG activity can have different properties, including: "rhythmicity," EEG activity is present in waves of approximately constant frequency; "arrhythmia," where there is no rhythmically-stabilized EEG activity; "dysrhythmia," a rhythm and/or pattern of EEG activity that rarely occurs or is seen in healthy subjects and that characteristically occurs in a patient group.

Hypothesis testing is the use of statistics to determine the probability that a given hypothesis is true.

"montage" means the placement of electrodes. EEG can be monitored using a bipolar montage or a reference montage. Bipolar means that there are two electrodes per channel, so there is a reference electrode for each channel. Reference montage means that there is a common reference electrode for all channels.

"morphology" refers to the shape of a waveform. The shape of an EEG pattern or wave is determined by the frequency of the combined constituent waveforms and their phase and voltage relationships. The wave mode can be described as: "singlet," which looks like a unique (distint) EEG activity consisting of one dominant activity; "polymorphism," a unique EEG activity consisting of multiple frequencies combined to form a complex waveform; "sinusoidal," a wave similar to a sine wave, with the singlet activity typically being sinusoidal; "transients," isolated waves or patterns that are significantly different from background activity.

"spike" refers to a transient having a peaked peak and a duration of from 20 to 70 milliseconds or less.

The term "spike" refers to a transient having a peaked peak and a duration of 70-200 milliseconds.

The term "neural network algorithm" refers to an algorithm that identifies sharp transients that are likely epileptiform abnormalities.

"noise" refers to any unwanted signal that modifies the desired signal. It may have multiple sources.

A null hypothesis is a hypothesis that is checked for possible rejections if the hypothesis is true (usually the observation is a casual result).

"periodic" refers to the distribution of patterns or elements over time (e.g., particular EEG activities occur at more or less regular intervals). The activities may be generalized, centralized, or lateralized.

The probability density P (x) of the continuous distribution is defined as the derivative of the distribution function D (x), where

D’(x)＝[P(x)]^X _-∞＝P(x)-P(-∞)＝P(x)

Thus, D (X) ═ P (X ≦ z ≦ X ≦ -_-∞ ^XP(y)dy

"sampling" or the term "sampling a signal" refers to reducing a continuous signal to a discrete signal. The digital signal is a sampled signal; obtained by sampling the analog signal at discrete points in time.

The term "sampling interval" is the time between successive samples; these points are typically evenly spaced in time.

The term "sampling rate" refers to the frequency in hertz (Hz) at which an analog-to-digital converter (ADC) samples an input analog signal.

The term "signal-to-noise ratio" (SNR) refers to a measure of the magnitude of the change in signal relative to the change in noise.

The EEG epoch (epoch) is the amplitude of the EEG signal, which is a function of time and frequency.

EEG recordings have a lot of information and it is difficult to determine whether the two sets of epochs are comparable or not.

Disclosure of Invention

The present invention provides a solution to the shortcomings of the prior art. The present invention provides a method and system for utilizing an empirical zero hypothesis for biological time series. The use of large-scale estimation methods facilitates the estimation of empirical zero density rather than using theoretical density. Empirical zeros are considered to be more dispersed than the usual theoretical zero distribution.

In one aspect of the invention, a method for determining whether two epochs are comparable is disclosed. The method comprises the step of selecting a first portion of a biological time series. The method then establishes the first portion as a baseline portion. Further, a second portion is selected for measurement and the similarity of the second portion to the portion of the base line is determined using an empirical zero assumption.

In another aspect of the invention, a method for determining whether two sets of epochs are comparable is disclosed. The method includes the steps of selecting a first plurality of epochs of the EEG recording and establishing them as baseline segments. The method further comprises the following steps: generating a second plurality of epochs; and estimating empirical zero hypotheses for a subset of epochs using a large-scale test to calculate an amount by which the second plurality of epochs are similar to the first plurality of epochs. The method also includes determining whether the second plurality of epochs are from an EEG recording.

In yet another aspect of the invention, an EEG system. The EEG system includes a processor communicatively coupled to a computer readable medium. The computer-readable medium includes instructions executable by the processor for: selecting a first epoch of an EEG recording; establishing a first time period as a baseline time period; selecting a second time period for measurement; and using an empirical zero assumption to determine the similarity of the second epoch to the baseline epoch.

In an alternative embodiment, the empirical zero value measures a value derived from the EEG. An empirical null may measure the amount of power in a given frequency range of the EEG.

The result is a Z-score expressing the difference from baseline, which can then be displayed in time series to show how the derived values vary from baseline over time.

Drawings

FIG. 1 is a system for utilizing an empirical zero hypothesis for biological time series.

FIG. 2 is a flow chart of a method utilizing an empirical zero hypothesis for a biological time series.

FIG. 3 is a flow chart of a method for utilizing an empirical zero hypothesis for a biological time series.

Fig. 4 is a block diagram of a computing device for EEG processing.

Fig. 5 is a graph illustrating an empirical zero hypothesis.

Fig. 6 is an illustration of an EEG system for use with a patient.

Figure 7 is a diagram of an international 10-20 electrode system showing electrode placement for EEG.

FIG. 8 is a detail drawing showing the middle 10% electrode position for EEG electrode placement as standardized by the American electroencephalogram society.

Fig. 9 is a graphical display of the amount of artifact present in an EEG recording.

Fig. 9A is a graphical display of the amount of artifact present in an EEG recording.

Fig. 9B is an enlarged isolated view of box 1B of the seizure probability channel of fig. 9.

Fig. 9C is an enlarged isolated view of a horizontal line of the artifact intensity channel of fig. 9.

FIG. 10 illustrates a variable window for a statistical automatic neural network application.

Fig. 11 illustrates an ANS window for a statistical automatic neural network application.

Fig. 12 is a block diagram of an MPL architecture.

Fig. 13 is an image of EEG activity.

Fig. 14 is an image of EEEG activity.

Detailed Description

Best mode(s) for carrying out the invention

The present invention provides a solution to the shortcomings of the prior art. The present invention provides a method and system for utilizing an empirical zero hypothesis for biological time series. More specifically, the present invention provides a method and system for utilizing an empirical zero hypothesis for raw scores generated by an artificial neural network.

Empirical assumptions refer to the use of working assumptions that can be examined using observation and experimentation, in other words, the assumptions are evidence-based. Empirical data was generated by experiment and observation. When researchers believe that there is no relationship or lack of information between two variables to state a scientific hypothesis, there will be a null hypothesis (by H)₀Coming labelShown). According to the null hypothesis, there was no significant difference between the specified populations, and any observed differences were due to sampling or experimental errors.

An empirical zero assumption is generally assumed to be true until evidence indicates the opposite. If the zero hypothesis is true, then the zero hypothesis is rejected if the observed data is significantly unlikely to occur. In this case, the null hypothesis is rejected and the alternative hypothesis is accepted to replace it. If the data is consistent with the null hypothesis, the null hypothesis is not rejected.

In the description provided below, the empirical zero hypothesis is explained as an example.

Hypothesis testing begins with the collection of zero hypotheses

H1，H2，...，HN

The corresponding test statistics, which may not be independent,

Y1，Y2，...，YN

and P-values P1, P2., PN, where Pi measures the observed value Yi of Yi negating the strength of Hi, e.g., Pi ═ probHi { | Yi > | }. By "large scale" is meant that N is a large number, at least greater than 100.

It is not necessary but convenient to use the value of z instead of the value of Yi or Pi,

zi＝Φ-1(Pi)，i＝1，2，...，N， (1.3)

Φ denotes a standard normal cumulative distribution function (cdf), Φ -1(.95) ═ 1.645, and the like. If Hi is completely true, zi will have a standard normal distribution

zi|Hi～N(0,1)。

This is referred to herein as the theoretical zero hypothesis.

Fig. 5 is a graph depicting an empirical zero hypothesis. Fig. 5 illustrates a histogram of 444 z-values, where a value of negative zi indicates a greater mutational effect. The smoothing curve f (z) is a natural spline with seven degrees of freedom, fitted to the histogram counts by Poisson regression. It emphasizes the central peak around z-0, presumably the majority of uninteresting drug site combinations with negligible mutational effects. Near its center, the peak is well described by a normal density with a mean of-0.35 and a standard deviation of 1.20, which will be referred to as the empirical zero hypothesis

zi|Hi～N(-0.35,1.202)

Large scale simultaneous hypothesis testing, where the number of cases exceeds, for example, 100, allows empirical estimation of the null hypothesis distribution. Empirical zeros may be more widely distributed (more scattered) than the theoretical zeros that are commonly used for single hypothesis testing. The choice between an empirical zero value and a theoretical zero value can greatly affect which cases are identified as "significant" or "of interest" rather than "zero value" or "not of interest," regardless of which simultaneous hypothesis testing method is used.

There are many possible reasons for the over-dispersion of the empirical zero value distribution, which results in empirical zero values being preferred for simultaneous verification: (ii) a hidden dependency; small effects that are, for the most part, real but not of interest; the scientific intent of large-scale testing is different from that of individual hypothesis testing. The latter is most often designed to reject zero hypotheses with high probability. Large-scale testing is often a more screening operation aimed at identifying a small fraction of cases of interest-presumably on the order of 10%.

Having explained the process of empirical zero hypothesis in more detail, the description provided below now explains the solution for utilizing empirical zero hypothesis for biological time series. More specifically, the present invention provides a method and system for utilizing an empirical zero hypothesis for raw scores generated by an artificial neural network.

As shown in fig. 1, a system 100 for utilizing an empirical zero hypothesis for a biological time series is generally specified. The system 100 preferably includes a source 70, a processor 75, and a display 80. The source 70 generates a digital input signal that is received by a processor 75 connected to the source 70. Processor 75 includes a computer readable medium 76 and an empirical zero hypothesis engine 77. The computer readable medium includes instructions executable by the processor to: selecting a first epoch of an EEG recording; establishing a first time period as a baseline time period; selecting a second time period for measurement; and an empirical zero assumption is used to determine the similarity of the second epoch to the baseline epoch. The analysis results are transmitted by the processor to a display device for the user.

In another aspect of the invention, a method may be used to determine whether two sets of epochs are comparable. The method includes the steps of selecting a first plurality of epochs of the EEG recording and establishing them as baseline segments. The method further comprises the following steps: a second plurality of epochs are generated and an empirical zero hypothesis for a subset of the second epochs is estimated using a large-scale test to calculate an amount by which the second plurality of epochs is similar to the first plurality of epochs. The method also includes determining whether the second plurality of epochs are from an EEG recording.

In yet another aspect of the present invention, an EEG system is disclosed. The EEG system includes a processor communicatively coupled to a computer readable medium. The computer readable medium includes instructions executable by a processor to: a first plurality of epochs of the EEG recording are selected and established as baseline segments. The method further comprises the following steps: a second plurality of epochs are generated and an empirical zero hypothesis for a subset of the second epochs is estimated using a large-scale test to calculate an amount by which the second plurality of epochs is similar to the first plurality of epochs. In the example of the system 100, a probability score of 30% is given to a one-hundred-second duration period, and the system determines whether thirty of the one-hundred periods are actually epileptic seizures. This occurs by calibrating fifty epochs to measure whether or not a seizure occurred in fifteen of the fifty epochs, thereby selecting the first plurality of epochs of the EEG recording to occur as baseline segments. The calibration will provide a probability value that will be validated against the remaining fifty epochs to estimate an empirical zero hypothesis using a large scale test to determine that the second plurality of epochs are similar to the first plurality of epochs. If fifteen of the remaining fifty pieces of evidence indicate a seizure, then the probability value is validated.

A system for utilizing an empirical zero hypothesis for a biological time series is generally designated 100 and preferably includes a source 70, a processor 75, and a display 80. At step 1, source 70 generates a digital input signal that is received by processor 75 at step 2. The processor is connected to a source 70. The processor 75 also includes a computer readable medium 76. The computer readable medium includes instructions executable by the processor to: selecting a first epoch of an EEG recording; establishing a first time period as a baseline time period; selecting a second time period for measurement; and an empirical zero assumption is used to determine the similarity of the second epoch to the baseline epoch. The analysis results are transmitted by the processor to a display device for the user.

The source 70 in fig. 1 may be any device that provides data to a processor. The source may be an EEG machine that provides real-time biological signals detected from human brain neural activity, or historical data stored in any data storage area (e.g., a database, computer-readable storage medium, directory, etc.). The data storage area may include a computer-readable storage medium, such as a hard disk, a non-volatile solid state memory device, or the like. The data store may provide data storage services, such as database storage services, directory services, and the like. A bio-signal is any signal in an organism that can be continuously measured and monitored. The term biosignal is generally used to refer to a bioelectric signal or a biosignal, but it may refer to both electrical and non-electrical signals. It is generally understood to refer to time-varying signals only.

The system 100 includes a processor 75. The processor may be a general purpose computer, computing device, microprocessor, microcontroller, or any other computing device with a computer-readable medium without departing from the scope of the invention. The computer readable medium includes a set of instructions for a processor to perform a method. This method will be described in more detail below.

The processor transmits the calculated result to a display device, which may be any electronic device with a display screen, such as a mobile phone, a PDA, a computer, an LCD, an LED, a TV screen, etc., without limiting the scope of the invention.

The processor is configured to perform specific functions based on a specific set of instructions provided by the computer readable medium, as will be explained with the help of the following examples.

Fig. 2 illustrates a flow chart of a preferred embodiment of the present invention. FIG. 2 illustrates a flow chart of a method 100, the method 100 for determining similarity of a second portion to a base portion using an empirical zero assumption. At step 101, a first portion of a biological time series is selected by a processor. The processor includes a computer readable medium comprising a set of computer readable instructions for performing the method. At step 102, a selected first portion of the biological time series is established as a baseline portion. Step 102 proceeds to step 103 where a second portion is selected for measurement. At step 104, the second portion is compared to the baseline portion using an empirical zero hypothesis to determine similarity to the baseline portion. Biological time series can be applied to EEG, intracranial pressure, heart rate, etc.

Fig. 3 shows a flow diagram of a method 200 according to another embodiment of the invention. FIG. 11 shows a flowchart of steps performed by a processor, according to another embodiment of the invention. At step 201, a first plurality of epochs of an EEG recording is selected by the processor. The processor includes a computer readable medium comprising a set of computer readable instructions for performing the method. At step 201, the selected first plurality of epochs of the EEG recording are also established as baseline segments. Step 201 proceeds to step 202, which generates a second plurality of epochs for the measurement. At step 203, empirical zero hypotheses for a subset of the second plurality of epochs are estimated using a large-scale test to calculate quantities of the second plurality of epochs that are similar to the first plurality of epochs. The method also includes determining whether the second plurality of epochs are from an EEG recording.

As shown in fig. 4, the EEG machine component 95 utilized to determine whether the second plurality of epochs are from EEG recordings is preferably a computer comprising a peripheral interface 325, an output/a/V326, a communication/NIC 327, a processor 328, a memory 329, and a storage 330. One skilled in the relevant art will recognize that the machine component 95 may include other components without departing from the scope and spirit of the present invention.

In accordance with a novel aspect of the present invention, the processor is configured to calibrate and verify raw scores processed through the artifact reduction filter and the neural network algorithm. The process of calibration and verification is performed based on an empirical zero assumption. For example, a hundred epochs of one-second duration with a probability score of 20% for a given seizure are extracted, and the system determines whether twenty of the one hundred epochs are actually seizures. This occurs by calibrating fifty epochs to measure whether or not a seizure occurred in ten of the fifty epochs, selecting the first portion of the biological time series and establishing it as the baseline portion. The calibration will provide a probability value that will be validated for the remaining fifty epochs (which is the second part for the measurement). The probability value is verified by using an empirical zero hypothesis to determine the similarity of the second portion to the base portion. This also allows the neural network to be trained to generate validated probability values. This also allows the neural network to be trained to generate validated probability values.

In another embodiment of the present invention, with respect to the system 100 of FIG. 1, a method may be used to determine whether two sets of epochs are comparable. The method includes the steps of selecting a first plurality of epochs of the EEG recording and establishing them as baseline segments. The method selects a first set of epochs and calibrates the probability values for the seizures. For example, a probability score for a given seizure is extracted as a hundred epochs of 20% of one-second duration. The method also includes generating/selecting a second plurality of epochs and estimating empirical zero hypotheses for a subset of the epochs using a large-scale test to calculate quantities of the second plurality of epochs that are similar to the first plurality of epochs. The method checks for similarity between the first set of epochs and the second set of epochs. Based on an empirical null hypothesis, whether the probability values for seizures for the second set of epochs are similar to the selected first set of epochs. Based on the calculation, it may be determined whether the second plurality of epochs are from an EEG recording.

Artificial Neural Networks (ANN) have been used to address various tasks in a number of areas that are difficult to address using common rule-based programming. The ANN may be learned and adapted through a learning algorithm. The types of ANN and the architecture of ANN differ mainly in learning methods.

A system for training a neural network for detecting artifacts in EEG recordings, comprising: a plurality of electrodes for generating a plurality of EEG signals; a processor connected to the plurality of electrodes to generate an EEG recording from the plurality of EEG signals; and a display connected to the processor for displaying the EEG recording. Preferably, the processor is configured to train the neural network to learn to generate a plurality of sub-concept outputs from the first plurality of inputs.

In classification, a task is to give a set of variables x, called attribute variables or inputs₁…x_nIn the case of (2), for a variable called a categorical variable, y is equal to x₀To be classified or to be output. A classifier h: x → y is a function that maps instances of x to values of y. The classifier is learned from a data set d consisting of samples on (x, y). The learning task involves finding a suitable bayesian network given the data set d on U. Let U be { x ═ x ₁,…,x_n}，n>1 is a set of variables.

Fig. 12 is a block diagram of an MPL architecture with inputs 280 and 285. The perceptron models the biological neurons as a mathematical function,

wherein the input value is calculatedThe weighted sum y of (a). The weight is

The following is a perceptron training algorithm for training an MLP with K outputs.

For i ═ 1, …, K

For j equal to 0, …, d

w_ij￥rand(-0.01,0.01)

Repetition of

For all x in random order^t，r^t)∈X

For i ═ 1, …, K

o_i￥0

For j equal to 0, …, d

o_i￥o_i+w_ijx^t _j

For i ═ 1, …, K

y_i￥exp(o_i)/∑_kexp(o_k) For i ═ 1, …, K

For j equal to 0, …, d

Until convergence

Where η is the learning factor.

The following is a back propagation algorithm for training an MLP with K outputs.

All v are_ihAnd w_hjInitialization is a rand (-0.01, 0.01) repeat

For all (x) in random order^t，r^t) E X1 for H

z_h￥sigmoid(w^T _hx^t)

For i 1

y_i＝v^T _iz

For i 1

Δv_i＝η(r^t _i-y^t _i)z

For H1

Δw_h＝η(∑_i(r^t _i-y^t _i)v_ih)z_h(1-z_h)x^tFor i 1

v_i￥v_i+Δv_i

For H1

w_h￥w_h+Δw_h

Until convergence.

Fig. 6 illustrates a system 25 in which a user interface is used for automatic artifact filtering of EEG in which empirical zero hypotheses for a subset of the second sub-epochs are estimated using a large scale test to calculate an amount by which the second plurality of epochs are similar to the first plurality of epochs in order to determine whether the second plurality of epochs are from an EEG recording. The patient 15 wears an electrode cap 31 consisting of a plurality of electrodes 35a-35c, which is attached to the patient's head, wherein the leads 38 are connected from the electrodes 35 to an EEG machine assembly 40, which EEG machine assembly 40 comprises an amplifier 42 for amplifying the signals to a computer 41 having a processor which is used to analyze the signals from the electrodes 35 and create an EEG recording 51, which EEG recording 51 can be viewed on a display 50. Buttons on the computer 41 allow multiple filters to be applied to remove multiple artifacts from the EEG and generate a clean EEG, either by a keyboard or touch screen buttons on the display 50. A more thorough description of electrodes utilized with the present invention is detailed by Wilson et al in U.S. Pat. No. 8112141, "Method And Device For Quick Press on EEG Electrode," the entire contents of which are incorporated herein by reference. The EEG is optimized for automatic artifact filtering. The EEG recording is then processed using a neural network algorithm to generate a processed EEG recording (raw score). The processor 41 is further configured to calibrate the raw scores to generate a probability value that an event has occurred, and then generate a display of the probability value with respect to time. Further, the processor 41 is configured to verify the probability value. The processor is also connected to a display for displaying the final output.

The electrodes 35a-35c may comprise a thin metal, such as a moderately-conductive alloy. In one embodiment, the electrodes 35a-35c comprise nitinol or other suitable shape memory material, metal, or alloy. An electrode comprising one or more of these materials, metals or alloys may be bendable or deformable, but may then move toward its original shape. For example, the legs of the electrode comprising the shape memory metal may be configured to bend upon application of a force (e.g., a force applied by a human or machine prior to insertion of the electrode into the patient's skin) and return toward an initial configuration upon release of the force (e.g., after the electrode has been inserted into the patient's skin). The legs may be bent away from the plane of the electrode body prior to insertion into the patient and may be returned towards the plane so that the legs are substantially parallel below the patient's skin after insertion, thereby ensuring the position and insertion of the electrodes.

In another embodiment, the electrodes 35a-35c comprise stainless steel. The legs 102 of the steel electrodes 35a-35c are bent at an angle (e.g., about 25, 30, 40, 45, or 50 degrees) and pushed into the skin with sufficient force to secure the electrodes 35a-35 c. In other embodiments, the electrodes may be formed of any suitable conductive material or combination of materials, wherein at least a portion of the electrodes are formed of a conductive material. For example, other embodiments may include non-conductive materials with a conductive outer layer material.

EEG is optimized for automatic artifact filtering. The EEG recording is then processed using a neural network algorithm to generate a processed EEG recording, which is analyzed for display.

Additional descriptions of Analyzing EEG recordings are set forth in Wilson et al, patent application Ser. No. 13/620855, "Method and System For Analyzing An EEG Recording," filed on 9, 15, 2012, the entire contents of which are incorporated herein by reference.

The patient has a plurality of electrodes attached to the patient's head, with leads from the electrodes connected to an amplifier for amplifying the signals to a processor that is used to analyze the signals from the electrodes and create an EEG recording. The brain produces different signals at different locations on the patient's head. A plurality of electrodes are placed on the head of the patient. The CZ site is located in the center. The number of electrodes determines the number of channels of the EEG. A greater number of channels results in a more detailed presentation of the patient's brain activity. Preferably, each amplifier 42 of the EEG machine assembly 40 corresponds to two electrodes 35 attached to the head of the patient 15. The output from the EEG machine assembly 40 is the difference in electrical activity detected by the two electrodes. The placement of each electrode is critical for EEG reporting because the closer the electrode pairs are to each other, the smaller the difference in brain waves recorded by the EEG machine component 40. Figure 7 is a diagram of an international 10-20 electrode system showing electrode placement for EEG. The term "electrode positioning" (10/20 system) refers to the standardized placement of scalp electrodes for classical EEG recordings. The nature of the system is a 10/20 range of percent distance between the Nasion-Inion and the fixation point. These points are labeled as frontal pole (Fp), middle lobe (C), parietal lobe (P), occipital lobe (O), and temporal lobe (T). The centerline electrode is labeled with a subscript z, which represents zero. Odd numbers are used as subscripts for points on the left hemisphere, while even numbers are used as subscripts for points on the right hemisphere. FIG. 8 is a detail drawing showing the middle 10% electrode position for EEG electrode placement as standardized by the American electroencephalogram society. A more thorough description of electrodes utilized with the present invention is detailed in Wilson et al, U.S. Pat. No. 8112141, "Method And device for Quick Press On EEG Electrode," which is incorporated herein by reference in its entirety.

Algorithms for removing artifacts from EEG typically use Blind Source Separation (BSS) algorithms such as CCA (canonical correlation analysis) and ICA (independent component analysis) to transform signals from a set of channels into a set of component waves or "sources".

In one example, an algorithm known as BSS-CCA is used to remove the effects of muscle activity from the EEG. Using algorithms on recorded montages often does not produce the best results. In this case, it is generally preferred to use a montage in which the reference electrode is one of the apical electrodes (such as the CZ in the international 10-20 standard). In this algorithm, the recorded montage is first transformed into a CZ reference montage before removing the artifacts. In case the signal at CZ indicates that it is not the best choice, then the algorithm will continue down the list of possible reference electrodes in order to find the appropriate one.

In one example of removing a snapshot (eye blink), a time period is first separated into multiple sources using BSS (blind source separation). Each source is then reconstructed as a recorded montage and then as a CZ reference montage deemed optimal for identifying snapshot-type artifacts. The CZ reference montage channel is examined by the neural network to determine if it is likely to be any snapshot. If so, the particular source is removed and the algorithm proceeds to the next source. However, if there is a problem with the CZ electrode, then another reference electrode will be selected for the source.

Fig. 9, 9A, 9B, and 9C illustrate graphical displays of the amount of artifact present in EEG recordings where a large scale test is used to estimate empirical zero hypotheses for a subset of epochs to calculate the amount by which the second plurality of epochs are similar to the first plurality of epochs in order to determine whether the second plurality of epochs are from EEG recordings. "artifacts" are electrical signals detected by the EEG along the scalp, but they originate from non-brain origins. There are patient-related artifacts (e.g., motion, perspiration, ECG, eye movement) and technical artifacts (50/60Hz artifact, cable movement, electrode paste related). Persyst artifact reduction (perspective artifact reduction) uses blind source separation to separate the original EEG signal into its underlying components. It then uses a set of high-level neural networks to identify components resulting from various types of artifact sources. In fig. 9, the artifact intensity channel 110 is shown as a series of horizontal lines 111. The plurality of horizontal lines 111 shown include horizontal line 112 for muscle artifacts, horizontal line 113 for chewing artifacts, horizontal line 114 for vertical eye movement artifacts, and horizontal line 115 for lateral eye movement artifacts. A person skilled in the relevant art will recognize that more or fewer horizontal lines may be used without departing from the scope and spirit of the invention.

Also shown in fig. 9 and 9A are seizure probability channel 120, rhythm spectrum-left hemisphere channel 130, rhythm spectrum-right hemisphere channel 140, FFT spectrum-left hemisphere channel 150, FFT spectrum-right hemisphere channel 160, asymmetric relative spectrum channel 170, asymmetric absolute index channel 180, EEG channel 190, and suppression ratio-left hemisphere and right hemisphere channel 200.

The rhythmic spectrogram allows one to see the evolution of a seizure in a single image. The rhythmic spectrogram measures the amount of rhythm present at each frequency in the EEG recording. The rhythm profile 130 for the left hemisphere and the rhythm profile 140 for the right hemisphere show different frequencies of rhythm components, with darker colors being more rhythmic. Seizures are detected as a sudden increase in the rhythm of the theta frequency. The rhythm spectrogram measures the amount of rhythm present in each frequency band in the recording. The rhythmic spectrogram shows the evolution of a seizure in a single image.

Seizure probability trend shows the calculated probability of seizure activity over time. Seizure probability trends show the duration of detected seizures and also suggest recording areas that may be below the seizure detection threshold but are still of interest. When displayed with other trends, the seizure probability trend provides a comprehensive view of the quantitative changes in the EEG. Seizure probability trends are determined by seizure detection algorithms, as indicated by bars 125.

The FFT spectrogram 150 for the left hemisphere and the FFT spectrogram 160 for the right hemisphere show the power of different frequencies in the same/different colors. During a seizure, there is a flame-like increase in power. The increase in bright color on both hemispheres indicates an increase in power at higher frequencies.

Fig. 9B shows a seizure 125 detected in seizure probability channel 120. Seizure probability trend shows the calculated probability of seizures over time, shows the duration of detected seizures, and also suggests recording areas that may be below the seizure detection threshold but are still of interest.

Fig. 9C shows the artifact intensity channels 110 as a series of horizontal lines 111. The plurality of horizontal lines 111 shown include horizontal line 112 for muscle artifacts, horizontal line 113 for chewing artifacts, horizontal line 114 for vertical eye movement artifacts, and horizontal line 115 for lateral eye movement artifacts. A person skilled in the relevant art will recognize that more or fewer horizontal lines may be used without departing from the scope and spirit of the invention.

Figure 10 illustrates the variable window of the statistical automated neural network application 3100.

Fig. 11 shows an ANS window 1150 for a statistical automatic neural network application. The window includes a network type form 270 and a training/retraining network form 275.

In a preferred embodiment, the empirical zero value measures a value derived from the EEG. An empirical null may measure the amount of power in a given frequency range of the EEG. The result is a Z-score expressing the difference from baseline, which can then be displayed in time series to show how the derived values vary from baseline over time. The four frequency bands of fig. 14 are graphs illustrating the z-scores of the derived values of the EEG (shown in fig. 13) for the FFTs included at the respective time, frequency and electrode groups. The image may be displayed on a display device as an output of an EEG. The images are used to detect epileptic seizures. The images depict a quantitative analysis of epochs received from an EEG recording machine. Starting from the top of the image, the highest trend shows the artifact strength. The artifact intensity channel shows a plurality of lines, which imply different types of artifacts, such as muscle artifacts, chewing artifacts, vertical eye movement artifacts, lateral eye movement artifacts, and the like. In other words, it shows how much artifact is present in the EEG at a particular point in time. The next trends are seizure detection and seizure probability. Seizure probability trend shows the calculated probability of seizure activity over time. Seizure probability trends show the duration of detected seizures and also suggest recording areas that may be below the seizure detection threshold but are still of interest. Since there are no seizures during this time period, they are blank in the provided image. Following seizure detection and seizure probability, the following trends are spike detection and spike rate. In addition, other trends may be displayed on the display screen. Examples for other trends may be FFT trends, asymmetric spectrograms and rejection ratios. The FFT trend is also referred to as CSA trend.

The implementation of the empirical zero hypothesis together with the neural network improves the efficiency of the quantitative analysis results displayed on the display screen. In light of the foregoing, it is believed that one skilled in the relevant art will recognize advantageous developments in the invention.

A more thorough description of electrodes utilized with the present invention is detailed in Wilson et al, U.S. Pat. No. 8112141, "Method And Device For Quick Press on EEG Electrode," which is incorporated herein by reference in its entirety. A more thorough description of EEG analysis utilized with the present invention is detailed in Wilson et al, U.S. patent application Ser. No. 13/620855, "method and System For Analyzing An EEG Recording," filed on 9, 15, 2012, the entire contents of which are incorporated herein by reference. A more thorough description of a user interface utilized with the present invention is detailed In Wilson et al, U.S. Pat. No. 9055927, "user interface For Artifact Removal In An EEG," which is incorporated herein by reference In its entirety. Additional description of analyzing EEG recordings is set forth in Wilson et al, U.S. patent application Ser. No. 13/684556, "Method And System For Detecting And Removing EEGArtifications," filed on 25/11/2012, the entire contents of which are incorporated herein by reference. A more thorough description of EEGs utilized in connection with the present invention is detailed in U.S. Pat. No. 8666484 "Method And System For Displaying EEGs" to Nierenberg et al, which is incorporated herein by reference in its entirety. A more thorough description of EEG recordings utilized with the present invention is detailed In Wilson et al, U.S. Pat. No. 9232922, "User Interface For artifact removal In An EEG," which is incorporated herein by reference In its entirety. Additional description of the qEEG is set forth in U.S. patent application Ser. No. 13/830742 "Method And System To calcium qEEG" To Nierenberg et al, filed on 3, 14, 2013, the entire contents of which are incorporated herein by reference. Additional description of the use of Neural networks with the present invention is set forth in Wilson, U.S. patent application Ser. No. 14/078497 "method and System Training A Neural Network", filed on 12.11.2013, the entire contents of which are incorporated herein by reference. Additional description of the use of neural networks with the present invention is set forth in U.S. patent application No. 14/222655, "System And Method For Generating a Proavailability Value For AnEvent," filed on 20.1.2014, to Nierenberg et al, the entire contents of which are incorporated herein by reference. U.S. patent application Ser. No. 15/131216 to Wilson et al, filed 2016, 4, 18, which is incorporated herein by reference in its entirety.