Processing for multi-channel signals

Abstract

Method and apparatus for improved processing for multi-channel signals. In an exemplary embodiment, an anomaly metric is computed for a multi-channel signal over a time window. The magnitude of the anomaly metric may be used to determine whether an anomaly is present in the multi-channel signal over the time window. In an exemplary embodiment, the anomaly metric may be a condition number associated with the singular values of the multi-channel signal over the time window, as further adjusted by the number of channels to produce a data condition number. Applications of the anomaly metric computation include the scrubbing of signal archives for epileptic seizure detection/prediction/counter-prediction algorithm training, pre-processing of multi-channel signals for real-time monitoring of bio-systems, and boot-up and/or adaptive self-checking of such systems during normal operation.

Description

TECHNICAL FIELD

The present disclosure relates generally to methods for processing multi-channel signals. In particular, the present disclosure relates to improved processing of multi-channel signals by detecting and/or treating possible anomalies in the multi-channel signals

INCORPORATION BY REFERENCE

All publications and patent applications mentioned in this specification are herein incorporated by reference to the same extent as if each individual publication or patent application was specifically and individually indicated to be incorporated by reference.

BACKGROUND OF THE INVENTION

A common task encountered in the field of signal processing is the sampling and processing of a physical state using multiple, ideally independent, signal sensors. The diversity of the resulting multi-sensor or multi-channel signal typically reveals more information about the underlying sampled state than can be obtained from employing a single sensor.

Multi-channel signal processing is utilized in biomedical applications. For example, in the field of neurological monitoring for epileptic seizure prediction, multiple electrodes may be implanted in diverse locations on or in a patient's brain to monitor the susceptibility of the patient to enter into an epileptic seizure. The multi-channel signals generated by the electrodes may be processed to, e.g., alert the patient and/or medical personnel of a high likelihood of imminent seizure. See, e.g., U.S. Patent Publication No. 2008/0183096 A1, “Systems and Methods for Identifying a Contra-ictal Condition in a Subject,” filed Jan. 25, 2008, assigned to the assignee of the present application, the contents of which are hereby incorporated by reference in their entirety. The signals may also be stored and processed offline to, e.g., train customized algorithms for estimating the likelihood that a patient will experience an imminent seizure. See, e.g., U.S. Pat. No. 6,678,548, “Unified probabilistic framework for predicting and detecting seizure onsets in the brain and multitherapeutic device,” the contents of which are hereby incorporated by reference in their entirety.

Other applications of multi-channel signal processing include the reception of wireless signals by a communications device using multiple antennas, geological monitoring of seismic activity for earthquake prediction, stereo imaging using multiple video cameras, etc.

When multi-channel signals are sampled over an extended period of time, artifacts or anomalies often appear in the signal. Such anomalies may be due to interference from external sources, disruptions to the power supply of the sensors, and/or other sources. Left untreated, such anomalies may degrade the quality of the measured signal and disrupt the accuracy of any subsequent processing of the multi-channel signal.

It would be desirable to have techniques to detect the presence of anomalies in a multi-channel signal, and to optimize the processing of a signal containing such anomalies.

SUMMARY OF THE INVENTION

An aspect of the present disclosure provides a method for displaying information associated with a multi-channel signal to a user, the method comprising: accepting input from the user selecting a metric to be displayed; displaying at least one time-series plot of the selected metric associated with the multi-channel signal; using a backdrop pattern, indicating on the at least one time-series plot portions of the plot having at least one identified characteristic; for each of the at least one time-series plot displayed, displaying a time event index wherein the corresponding metric meets a predetermined condition; accepting input from the user as to whether to display further information associated with the time event index; and displaying the further information associated with the time event index if the user so specifies.

Another aspect of the present disclosure provides a method for detecting anomalies in a multi-channel signal, the method comprising: sampling the multi-channel signal over a time window; computing an anomaly metric for the multi-channel signal over the time window; and identifying the presence of an anomaly based on the magnitude of the anomaly metric; the computing an anomaly metric comprising: computing a condition number of the multi-channel signal over the time window; and adjusting the condition number based on a parameter of the multi-channel signal to generate a data condition number (DCN); the identifying the presence of an anomaly comprising comparing the magnitude of the data condition number (DCN) to at least one threshold; the method further comprising generating one of the at least one threshold, the generating comprising: generating a histogram of number of anomalies detected for each of a plurality of candidate thresholds; generating a cumulative distribution function from the histogram, the cumulative distribution function mapping each candidate threshold to a percentage value; and determining the generated threshold as the candidate threshold mapped to a corresponding predetermined percentage value by the cumulative distribution function.

Yet another aspect of the present disclosure provides a method for detecting anomalies in a multi-channel signal, the method comprising: sampling the multi-channel signal over a time window; computing an anomaly metric for the multi-channel signal over the time window; and identifying the presence of an anomaly based on the magnitude of the anomaly metric; the computing an anomaly metric comprising: computing a condition number of the multi-channel signal over the time window; and adjusting the condition number based on a parameter of the multi-channel signal to generate a data condition number (DCN); the identifying the presence of an anomaly comprising comparing the magnitude of the rate of change of the data condition number (DCN) to at least one threshold.

Yet another aspect of the present disclosure provides a method for detecting anomalies in a multi-channel signal, the method comprising: sampling the multi-channel signal over a time window; computing an anomaly metric for the multi-channel signal over the time window; and identifying the presence of an anomaly based on the magnitude of the anomaly metric; the computing an anomaly metric comprising: computing a condition number of the multi-channel signal over the time window; and adjusting the condition number based on a parameter of the multi-channel signal to generate a data condition number (DCN); the multi-channel signal comprising a signal sampled from a plurality of electrodes implanted on or in a patient's brain, the method further comprising: generating a DCN time series corresponding to a plurality of time windows; generating an anomaly log based on the DCN time series; merging anomalies in the anomaly log separated by less than a minimum separation to generate a modified anomaly log; identifying segments of the multi-channel signal corresponding to anomalies in the modified anomaly log; and outputting time-expanded versions of the identified segments to a record; the identifying the presence of an anomaly comprising matching the DCN time series to at least one known pattern of DCN time series corresponding to an anomalous condition.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 depicts an exemplary embodiment of the present disclosure of a system for quality management of measured electrical signals generated by electrodes placed on or within the brain.

FIGS. 2A, 2B and 2C illustrate possible signal anomalies that may be detected by the anomaly detector/processor in the multi-channel signal.

FIG. 3A depicts an exemplary method according to the present disclosure, wherein a metric known as a “data condition number,” or DCN, is derived for a given multi-channel signal.

FIG. 3B shows a matrix A[k] derived from the multi-channel signal.

FIG. 4A illustrates a data condition number (DCN) time series obtained from a sample multi-channel signal using the steps of FIG. 3A and thresholding for event detection.

FIG. 4B illustrates a plot used in a procedure by which specific thresholds may be automatically chosen for any patient or group of patients.

FIGS. 4C and 4D illustrate an alternative exemplary embodiment of a technique for choosing an optimum threshold T1*.

FIG. 5 depicts an exemplary method according to the present disclosure for taking action in response to the detection of an anomaly.

FIG. 6 depicts an alternative exemplary embodiment of the present disclosure, wherein the techniques disclosed hereinabove are applied in the context of a real-time patient monitoring and neurological event detection system.

FIG. 7 depicts a generalized block diagram of the real-time analysis system featuring an anomaly pre-processing block according to the present disclosure.

FIG. 8 depicts an exemplary method according to the present disclosure that may be implemented, e.g., using the system shown in FIG. 7.

FIGS. 9A and 9B illustrate exemplary plots of data condition number (DCN) versus time, corresponding to certain specific operating scenarios.

FIGS. 10A and 10B illustrate an exemplary embodiment of a graphical display interface for conveniently displaying information associated with the techniques of the present disclosure.

DETAILED DESCRIPTION OF THE INVENTION

The detailed description set forth below in connection with the appended drawings is intended as a description of exemplary embodiments of the present invention and is not intended to represent the only exemplary embodiments in which the present invention can be practiced. The term “exemplary” used throughout this description means “serving as an example, instance, or illustration,” and should not necessarily be construed as preferred or advantageous over other exemplary embodiments. The detailed description includes specific details for the purpose of providing a thorough understanding of the exemplary embodiments of the invention. It will be apparent to those skilled in the art that the exemplary embodiments of the invention may be practiced without these specific details. In some instances, well known structures and devices are shown in block diagram form in order to avoid obscuring the novelty of the exemplary embodiments presented herein.

While the discussion below focuses on measuring electrical signals generated by electrodes placed near, on, or within the brain or nervous system (EEG signals) of subjects and subject populations for the determination of when an epileptic subject is in a condition susceptible to seizure, it should be appreciated that the techniques of the present disclosure are not limited to measuring EEG signals or to determining when the subject is susceptible to seizure. For example, the techniques could also be used in systems that measure one or more of a blood pressure, blood oxygenation (e.g., via pulse oximetry), temperature of the brain or of portions of the subject, blood flow measurements, ECG/EKG, heart rate signals, respiratory signals, chemical concentrations of neurotransmitters, chemical concentrations of medications, pH in the blood, or other physiological or biochemical parameters of a subject.

The present disclosure may also be applicable to monitoring other neurological or psychiatric disorders and identifying a condition or state for such disorders in which the subject is unlikely to experience some adverse effect. For example, the present disclosure may also be applicable to monitoring and management of sleep apnea, Parkinson's disease, essential tremor, Alzheimer's disease, migraine headaches, depression, eating disorders, cardiac arrhythmias, bipolar spectrum disorders, or the like.

Non-biomedical applications of the techniques described herein are also contemplated to be within the scope of the present disclosure.

FIG. 1 depicts an exemplary embodiment of the present disclosure of a system for quality management of measured electrical signals generated by electrodes implanted into the brain, subdurally, epidurally, partially or fully in the skull, between the skull and one or more layers of the patient's scalp, or on the exterior of the patient's head. In some embodiments, the electrodes are configured to be inserted through a single opening in a skull of a patient and to be dispersed on a surface of the brain. The surface is preferably the cortical surface of the brain, but may alternatively be any other suitable surface or layer of the brain, such as the dura mater. Note the exemplary embodiment depicted in FIG. 1 is shown for illustrative purposes only, and is not meant to limit the scope of the present disclosure to any particular embodiment shown.

In FIG. 1, electrodes 110 are implanted on or in the brain of a patient 100, e.g., underneath the dura mater and on a cortical surface of the patient's brain. Each of the electrodes generates a corresponding signal that is input to a data pre-processing module 120. The electrodes may be directly connected to the data pre-processing module 120 through a series of electrical leads 115a, or they may be wirelessly connected to the data pre-processing module 120 over a wireless link. In an exemplary embodiment, data pre-processing module 120 may, e.g., digitize the plurality of received electrical signals 115b to generate a multi-channel signal 120a for further processing.

Multi-channel signal 120a is input to an anomaly detector/processor 130. In an exemplary embodiment, the anomaly detector/processor 130 may utilize techniques further disclosed hereinbelow to identify the presence of signal anomalies in the multi-channel signal 120a. As further disclosed hereinbelow, anomaly detector/processor 130 may also take further action to address the anomalies detected, e.g., flagging the portions of the multi-channel signal corresponding to the detected anomalies in a log file 130a for optional manual review by a human operator.

In the exemplary embodiment shown, the log file 130a from the anomaly detection/processing module 130 is provided along with the multi-channel signal 120a to a data processing/adaptive algorithm training module 140. In an exemplary embodiment, module 140 may utilize the multi-channel signal 120a, coupled with information from the log file 130a about which portions of the multi-channel signal 120a contain anomalies, to train an adaptive algorithm to identify conditions under which patient 100 is susceptible to seizure. An exemplary system is described in Snyder, et al., “The statistics of a practical seizure warning system,” Journal of Neural Engineering, vol. 5, pp. 392-401 (2008), the contents of which are incorporated by referenced herein in its entirety. In an exemplary embodiment, module 140 may, e.g., automatically de-emphasize portions of multi-channel signal 120a corresponding to signal anomalies, and emphasize other portions of the signal 120a, to configure adaptive weights for a seizure prediction algorithm 140a. In alternative exemplary embodiments, a human operator may manually review portions of multi-channel signal 120a that have been flagged in the log file 130a, and decide whether such portions may be used for adaptive algorithm training.

In an exemplary embodiment, the log file 130a need not be limited to a single file residing in a single piece of storage hardware. For example, the log file 130a can be an extensive archive of intracranial EEG patterns that can be used to develop a predictive neurosensing device for managing seizures by mining the archive for signal patterns over the patient population. This archive may be stored for multi-user access and processing in, e.g., a server or cloud computing system.

FIGS. 2A, 2B, and 2C illustrate possible signal anomalies that may be detected by the anomaly detector/processor 130 in the multi-channel signal 120a. Note the anomaly patterns identified in FIGS. 2A, 2B, and 2C are for illustrative purposes only, and are not meant to limit the scope of the present disclosure to any particular anomaly patterns highlighted.

In FIG. 2A, reference numeral 210 points to an instance of a “spikes without phase reversal” anomaly in a referential multi-channel signal, or “spikes without double-phase reversal” anomaly in a bipolar multi-channel signal, possibly due to non-physiological interference sources. Reference numeral 220 points to an instance of a “flatline” anomaly in the multi-channel signal. In FIG. 2B, reference numeral 230 points to an instance of a “line noise” anomaly in the multi-channel signal. Such an anomaly may correspond to, e.g., powerline noise, i.e., 50 or 60 Hz noise and/or harmonics superimposed on the multi-channel signal. In FIG. 2C, reference numeral 240 points to an instance of a “saturation” anomaly in the multi-channel signal.

Other possible anomalies in a multi-channel signal (not shown) include episodic artifacts such as motion (large swings in the multi-channel signal), DC shifts (different DC levels between different channels or across a single channel), pops (exponential decay from amplifier highpass characteristic of a sudden change in the DC level of a signal), and glitches (e.g., 50 ms burst transients in the signal). Long-term anomalies may include deterioration trends in the system, and/or channels of persistently poor quality. Such anomalies and others not explicitly enumerated are contemplated to be within the scope of the present disclosure.

FIG. 3A depicts an exemplary method 300 according to the present disclosure, wherein a specific anomaly metric known as a “data condition number,” or DCN, is derived for a given multi-channel signal. In an exemplary embodiment, the DCN may be used to help identify the presence of anomalies in the multi-channel signal. The DCN as described with reference to FIG. 3A has been found to provide a reliable indicator for the presence of anomalies in a multi-channel signal, and is readily applicable to a wide variety of scenarios.

In FIG. 3A, at step 310, the number of dimensions of a multi-channel signal is defined as the variable N. The number of dimensions may correspond to the total number of sensors, e.g., the number of independent electrodes placed in a patient's brain in the epilepsy monitoring system of FIG. 1.

At step 320, the multi-channel signal is divided in time into windows of duration T, with each of the time windows being indexed by a counter k. According to the present disclosure, the multi-channel signal may be a discrete-time signal sampled at a rate of S Hz. In this case, each time window k may contain a total of T·S discrete-time samples multiplied by N channels (or sensors), which may be arranged to form a T·S matrix by N A[k] at step 330. The matrix A[k] is also shown in FIG. 3B. In an exemplary embodiment, the matrix A[k] is generally rectangular.

In an exemplary embodiment, T may be 5 seconds, S may be 400 Hz, and N may be 128 for an epilepsy monitoring unit such as depicted in FIG. 1. In an alternative exemplary embodiment, such as the real-time patient monitoring and neurological event detection system 60 of FIG. 6 utilizing implanted electrodes, N may be 16.

In an exemplary embodiment, the time windows k may be chosen to collectively span the entire duration of the multi-channel signal, i.e., the time windows are non-overlapping and contiguous in time. In alternative exemplary embodiments, the time windows need not be contiguous in time, and may be spread out over the duration of the multi-channel signal. In this way, the time windows effectively sub-sample the total duration of the multi-channel signal. This sub-sampling may result in fewer matrices A[k] to be processed as compared to using contiguous time windows, reducing the computational complexity. In yet other alternative exemplary embodiments, the time windows also may be made overlapping in time. In an exemplary embodiment, the signals for the channels may be represented using a referential montage, wherein each signal is measured as an electrical potential with respect to a common (“ground”) contact placed somewhere else in the body; e.g., an “earlobe” in a scalp electroencephalogram.

At step 340, a condition number C[k] is computed for each matrix A[k] as follows:

$\begin{array}{cc}C\left[k\right]=\frac{\max \left\{{\sigma }_{ik}\right\}}{\min \left\{{\sigma }_{ik}\right\}};& \left(\mathrm{Equation}1\right)\end{array}$ wherein {σ_ik} is the set of singular values of each matrix A[k]. In an exemplary embodiment, the singular values of each matrix A[k] may be computed using a singular-value decomposition (SVD) well-known in the art:A[k]=USV^T;  (Equation 2)wherein U and V are both square unitary matrices, and S contains the singular values of A[k]. One of ordinary skill in the art will appreciate that a variety of software tools are available to calculate the singular values, or approximate singular values, of a given matrix, for the purpose of deriving the matrix S. Such software tools, include, e.g., the publicly available software packages LAPACK or EISPACK.

At step 350, the condition number C[k] is further refined by conversion into a “data condition number” DCN[k]. In an exemplary embodiment, DCN[k] may be computed as:

$\begin{array}{cc}\mathrm{DCN}\left[k\right]=1+\frac{\left(C\left[k\right]-1\right)}{N}.& \left(\mathrm{Equation}3\right)\end{array}$

The conversion from a condition number into a data condition number may be done to compensate for an expected quasi-linear increase in the condition number due to the number of channels N. Following the conversion, the data condition number may generally be processed independently of the number of channels. One of ordinary skill in the art will appreciate that alternative techniques for accounting for the number of channels may be employed. For example, the condition number need not be converted to a data condition number, and the thresholds used to determine the presence of anomalies may instead be adjusted by the number of channels. Such alternative exemplary embodiments are also contemplated to be within the scope of the present disclosure.

One of ordinary skill in the art will further appreciate that, in alternative exemplary embodiments, either the condition number C[k] or thresholds used may be alternatively, or further, normalized with respect to any source of variation in C[k] that is not of interest, i.e., not indicative of an anomaly. For example, variables such as the window size, signal measurement bandwidth, electrode montage, etc., may also be accounted for in converting the condition number C[k] to the data condition number DCN[k], and/or choosing the thresholds against which the DCN[k] is compared. Such additional coefficients and parameters may be calibrated empirically, and one of ordinary skill in the art may readily modify Equation 3 accordingly to account for such coefficients and parameters.

According to the present disclosure, the magnitude of the data condition number DCN[k] may serve as an indicator of whether a time window k of a multi-channel signal contains anomalies. For example, the DCN[k] may range from a value of 1, indicative of “healthy” data that is lacking in anomalies, to an arbitrarily large value ∞, indicative of “ill” data corresponding, e.g., to complete flatlining over the time window of interest. Evaluating the magnitude of each DCN[k] may provide an indication of whether a signal anomaly is present in the corresponding time window k.

In alternative exemplary embodiments (not shown), the DCN may be computed for a subset of the total number of signal sensors by constructing the matrix A[k] using such subset of signals, and employing the number of signals in the subset for the variable N. For example, instead of employing all N channels to construct the matrix A[k], only the signals from a subset N−1 of the channels may be used. This may be advantageous when, e.g., one of the signal sensors is known to be faulty. Such exemplary embodiments are contemplated to be within the scope of the present disclosure.

In alternative exemplary embodiments, the DCN may be computed from a subset N−1 of the channels to account for the effect of average montages, circular bipolar montages, or any other electrode montages, wherein one channel is known a priori to be a linear combination of all remaining channels, and therefore the matrix A[k] is rank-deficient.

One of ordinary skill in the art will appreciate that in light of the present disclosure, various alternative anomaly metrics to the condition number for detecting the presence of anomalies may be derived, based on, e.g., determining the independence or correlation between two or more of the channels. These alternative metrics may be derived based on the assumption that healthy data, especially in neurological multi-channel signals, is associated with independence among the channels, while data containing anomalies is associated with a lack of independence among the channels.

For example, for a square matrix A[k], a generalized condition number C′[k] may be derived as:C′[k]=Norm(A[k])·Norm(A⁻¹[k]);  (Equation 4)wherein Norm(·) denotes a norm of the matrix in parentheses, and A⁻[k] is the inverse of the square matrix A[k]. For example, norms such as the L1-norm and L-infinity norm are well-known in the art, and may be applied to compute an anomaly metric associated with a matrix A[k] in the following manner. One of ordinary skill in the art will appreciate that the L1-norm may be defined as the maximum of the column sums of A[k], and the L-infinity norm may be defined as the maximum of the row sums.

As the computation of the generalized condition number C′ [k] and other types of condition numbers may require that the matrix A[k] be, e.g., a square matrix, or have other pre-specified dimensions, the data from the multi-channel signals may be suitably modified to ensure that the matrix A[k] takes on the proper form. For example, to ensure that the matrix A[k] is a square matrix, the size of the time window may be chosen such that the number of discrete time samples for each channel is equal to the total number of channels. Alternatively, the discrete time samples over a given time window may be sub-sampled at regular intervals to arrive at a square matrix A[k]. Such exemplary embodiments are contemplated to be within the scope of the present disclosure.

FIG. 4A illustrates a data condition number time series 400 obtained from a sample multi-channel signal using the steps of FIG. 3A and thresholding for event detection. In FIG. 4A, the horizontal axis depicts the time window index k, while the vertical axis depicts the value of DCN[k] corresponding to k. Note the absolute values of DCN[k] shown in FIG. 4A and discussed below are given for illustration only, and are not meant to restrict the techniques of the present disclosure to any particular values of DCN[k] shown. In particular, the absolute values of DCN[k] may depend on various parameters of the application.

In FIG. 4A, DCN[k] is shown to take on values generally less than 500 prior to time window k=K1. Between k=K1 and k=K2, DCN[k] is approximately 1000, while at time k=K2, DCN[k] is greater than 2000. In an exemplary embodiment, one or more predetermined thresholds may be chosen to identify the presence of anomalies in the multi-channel signal. For example, in FIG. 4A, the DCN[k] may be compared to a single threshold T1, and values of DCN[k] that are greater than the threshold T1 may indicate the presence of an “event” in the corresponding time window k. In an exemplary embodiment, the presence of an event may be taken to indicate the presence of an anomaly.

Further shown in FIG. 4A is an anomaly plot 410 generated from the time series 400 by comparison with threshold T1. Note anomaly plot 410 identifies the presence of two events A and B, possibly corresponding to two anomalies.

In alternative exemplary embodiments, two or more thresholds may be chosen for more precise categorization of the DCN. For example, in an exemplary embodiment, two thresholds T1 and T2 may be chosen, wherein T1<T2. In this exemplary embodiment, if the DCN is less than T1, then a lack of anomaly in the time window may be declared. If the DCN is greater than T2, then an anomaly can be automatically declared. If the DCN is between T1 and T2, then further processing, such as manual inspection of the multi-channel signal, may be performed to determine whether an anomaly is actually present. In an exemplary embodiment, patient-specific thresholds may be chosen that are customized to an individual patient whose neurological or other biological state is being monitored by the multiple sensors. Thresholds may be set differently for different patients, to account for the unique characteristics of each patient's bio-signals.

According to an aspect of the present disclosure, specific thresholds may be automatically chosen for any patient or group of patients using a procedure as described with reference to the plot 400A shown in FIG. 4B. In FIG. 4B, the horizontal axis shows a candidate threshold T1, while the vertical axis tallies a number of events detected for the corresponding candidate threshold. Based on the plot 400A, an optimum threshold T1* may be chosen to, e.g., maximize the probability of detecting an anomalous event while minimizing the probability of “false” positives. In an exemplary embodiment, as shown in FIG. 4B, such an optimum threshold T1* may be chosen as, e.g., an inflection point of the plot 400A having the largest abscissa from amongst all inflection points, wherein an inflection is defined as a point where the curvature changes sign, e.g., from facing down to facing up. Alternatively, if such an inflection point does not exist (e.g., in exponential decay), the threshold T1* may be chosen as the minimum abscissa where the plot 400A drops to between 1% and 2% of its maximum value Nmax.

FIGS. 4C and 4D illustrate an alternative exemplary embodiment of a technique for choosing an optimum threshold T1*. FIG. 4C illustrates an exemplary histogram of DCN (horizontal axis), with the vertical axis showing the number of instances of the corresponding DCN in an arbitrary multi-channel signal (not shown). From the histogram of FIG. 4C, one of ordinary skill in the art may readily derive a corresponding cumulative distribution function (CDF), as shown in FIG. 4D, which indicates what percentage of events correspond to DCN's equal to or less than the shown DCN. One of ordinary skill in the art will appreciate that by selecting an appropriate percentage (e.g., 90% as illustrated in FIG. 4D), a corresponding threshold DCN of T1*, such as T1*_90% in FIG. 4D, may be automatically determined using the functional mapping provided by the CDF. One of ordinary skill in the art will further appreciate that in an exemplary embodiment, a computer may be programmed to perform such threshold selection, such that the choice of threshold may be made fully automatic without the need for manual intervention.

While certain exemplary techniques for automatically a choosing a suitable threshold T1* have been disclosed hereinabove, one of ordinary skill in the art will appreciate that alternative techniques not explicitly described may be readily derived in light of the present disclosure. Such alternative exemplary embodiments are contemplated to be within the scope of the present disclosure.

In alternative exemplary embodiments, additional properties of the DCN may be analyzed to further aid in the detection of anomalies in the multi-channel signal. For example, the rate of change of the DCN over a predetermined interval of time may also be utilized to detect the presence of an anomaly. Such modifications to the DCN and others not explicitly described will be clear to one of ordinary skill in the art, and are contemplated to be within the scope of the present disclosure.

FIG. 5 depicts an exemplary method according to the present disclosure for taking action in response to the detection of an anomaly.

At step 510, a DCN time series such as 400 in FIG. 4A is generated from a multi-channel signal. In an exemplary embodiment, such a time series may be generated according to the method 300 depicted in FIG. 3A. However, one of ordinary skill in the art will appreciate that alternative methods may be employed to generate a suitable time series in light of the present disclosure.

At step 520, an anomaly log is generated based on the DCN time series. Such an anomaly log may identify, e.g., time window indices k in the DCN time series corresponding to detected anomalies. For example, in the exemplary embodiment wherein DCN[k] is compared to a single threshold T1 to determine the presence of an anomaly, the anomaly log may record all time window indices k wherein DCN[k] is larger than T1. As such, the anomaly log may effectively capture the relevant information from the anomaly plot 410.

One of ordinary skill in the art will appreciate that the information in an anomaly log may be recorded in several ways. For example, each line in the anomaly log may record the time index k associated with the beginning of a detected anomaly, and the corresponding time duration of the detected anomaly. Alternatively, the start and stop time indices associated with each detected anomaly may be recorded. Such exemplary embodiments are contemplated to be within the scope of the present disclosure.

At step 530, the complexity of the anomaly log may be reduced by merging separate anomalies that are separated by less than a minimum time separation. For example, assume two anomalies each of duration 10 are found in an anomaly log starting at time windows k=1 and k=12, i.e., the two anomalies are separated by a time duration of Δk=1. If a minimum time separation is defined as Δk_min=5, then the two anomalies may be merged to form a single anomaly, which can be recorded in a simplified anomaly log as a single merged anomaly of duration Δk=21 starting at k=1.

By performing the merging as described at step 530, the number of recorded anomalies and the size of the resulting anomaly log may be reduced to facilitate subsequent processing.

At step 540, the segments of the original multi-channel signal corresponding to the detected anomalies are identified.

At step 550, the identified segments of the multi-channel signal may be stored in an output record for post-processing. For example, the output record may be a computer file stored in a storage medium such as a computer hard drive, or it may be a paper print-out. In an exemplary embodiment, the identified segments output to the file may be expanded beyond those strictly associated with the anomalies. For example, fixed time segments of the multi-channel signal both immediately prior to and immediately subsequent to each identified data anomaly may also be output for each identified segment corresponding to an anomaly. The additional segments may further aid in the post-processing of the anomalies in the multi-channel signal, as further described hereinbelow.

In an exemplary embodiment (not shown), the output record generated by the method 500 may be manually reviewed, or “scrubbed,” by a human technician to verify the presence of anomalies in the identified multi-channel signal segments. If the identified segment is verified to contain an anomaly, the segment may be, e.g., omitted from further post-processing, or other measures may be taken.

FIG. 6 depicts an alternative exemplary embodiment of the present disclosure, wherein the techniques disclosed hereinabove are applied in the context of a real-time patient monitoring and neurological event detection system 60. For a more detailed description of the system in FIG. 6, see, e.g., “Minimally Invasive Monitoring Methods,” U.S. patent application Ser. No. 11/766,751, filed Jun. 21, 2007, published as U.S. Patent Publication No. 2008/0027347A1 and assigned to the assignee of the present application, the contents of which are hereby incorporated by reference in their entirety. Note that FIG. 6 is provided for illustrative purposes only, and is not meant to limit the scope of the present disclosure in any way.

In FIG. 6, system 60 includes one or more implantable sensors or devices 62 that are configured to sample electrical activity from the patient's brain (e.g., EEG signals). The implantable devices may be active (with internal power source), passive (no internal power source), or semi-passive (internal power source to power components, but not to transmit data signal). The implantable devices 62 may be implanted anywhere in the patient. In an exemplary embodiment, one or more of the devices 62 may be implanted adjacent to a previously identified epileptic focus or a portion of the brain where the focus is believed to be located. Alternatively, the devices 62 themselves may be used to help determine the location of an epileptic focus

In one aspect, the neural signals of the patient are sampled substantially continuously with the electrodes coupled to the electronic components of the implanted leadless device. A wireless signal is transmitted that is encoded with data that is indicative of the sampled neural signal from the implanted device to an external device. The wireless signal that is encoded with data that is indicative of the sampled neural signal is derived from the wireless signal received from the external device. The wireless signal can be any type of wireless signal—radiofrequency signal, magnetic signal, optical signal, acoustic signal, infrared signal, or the like.

The physician may implant any desired number of devices in the patient. As noted above, in addition to monitoring brain signals, one or more additional implanted devices 62 may be implanted to measure other physiological signals from the patient.

Implantable devices 62 may be configured to substantially continuously sample the brain activity of the groups of neurons in the immediate vicinity of the implanted device. The implantable devices 62 may be interrogated and powered by a signal from an external device 64 to facilitate the substantially continuous sampling of the brain activity signals. Sampling of the brain activity may be carried out at a rate above about 200 Hz, and preferably between about 200 Hz and about 1000 Hz, and most preferably at about 400 Hz, but it could be higher or lower, depending on the specific condition being monitored, the patient, and other factors. Each sample of the patient's brain activity may contain between about 8 bits per sample and about 32 bits per sample, and preferably between about 12 bits per sample and about 16 bits per sample.

In alternative embodiments, it may be desirable to have the implantable devices sample the brain activity of the patient on a non-continuous basis. In such embodiments, the implantable devices 62 may be configured to sample the brain activity signals periodically (e.g., once every 10 seconds) or aperiodically.

Implantable devices 16 may comprise a separate memory module for storing the recorded brain activity signals, a unique identification code for the device, algorithms, other programming, or the like.

A patient instrumented with the implanted devices 62 may carry a data collection device 64 that is external to the patient's body. The external device 64 would receive and store the signals from the implanted devices 62 with the encoded EEG data (or other physiological signals). The signals received from the plurality of implanted devices 62 may be represented as a multi-channel signal, and may be pre-processed according to the techniques of the present disclosure. The external device 64 is typically of a size so as to be portable and carried by the patient in a pocket or bag that is maintained in close proximity to the patient. In alternative embodiments, the device may be configured to be used in a hospital setting and placed alongside a patient's bed. Communication between the data collection device 64 and the implantable device 62 may take place through wireless communication. The wireless communication link between implantable device 62 and external device 64 may provide a communication link for transmitting data and/or power. External device 64 may include a control module 66 that communicates with the implanted device through an antenna 68. In the illustrated embodiment, antenna 68 is in the form of a necklace that is in communication range with the implantable devices 62.

Transmission of data and power between implantable device 62 and external device 64 may be carried out through a radiofrequency link, infrared link, magnetic induction, electromagnetic link, Bluetooth® link, Zigbee link, sonic link, optical link, other types of wireless links, or combinations thereof.

In an exemplary embodiment, the external device 64 may include software to pre-process the data according to the present disclosure and analyze the data in substantially real-time. For example, the received RF signal with the sampled EEG may be analyzed for the presence of anomalies according to the present disclosure, and further by EEG analysis algorithms to estimate the patient's brain state which is typically indicative of the patient's propensity for a neurological event. The neurological event may be a seizure, migraine headache, episode of depression, tremor, or the like. The estimation of the patient's brain state may cause generation of an output. The output may be in the form of a control signal to activate a therapeutic device (e.g., implanted in the patient, such as a vagus nerve stimulator, deep brain or cortical stimulator, implanted drug pump, etc.).

In an exemplary embodiment, the output may be used to activate a user interface on the external device to produce an output communication to the patient. For example, the external device may be used to provide a substantially continuous output or periodic output communication to the patient that indicates their brain state and/or propensity for the neurological event. Such a communication could allow the patient to manually initiate self-therapy (e.g., wave wand over implanted vagus nerve stimulator, cortical, or deep brain stimulator, take a fast acting anti-epileptic drug, etc.).

In an alternative exemplary embodiment, the external device 64 may further communicate with an auxiliary server (not shown) having more extensive computational and storage resources than can be supported in the form factor of the external device 64. In such an exemplary embodiment, the anomaly pre-processing and EEG analysis algorithms may be performed by an auxiliary server, or the computations of the external device 64 may be otherwise facilitated by the computational resources of the auxiliary server.

FIG. 7 depicts a generalized block diagram 700 of the real-time analysis system 60 featuring an anomaly pre-processing block 735 according to the present disclosure. One of ordinary skill in the art will appreciate that the block diagram 700 need not be limited to the exemplary system 60 shown in FIG. 6, but may also be broadly applicable to other types of multi-channel sensing and processing systems.

In FIG. 7, biosensor signals 710 form a multi-channel signal that is provided over a wireless link to wireless unit 720. Wireless unit 720 communicates the multi-channel signal to a processing module 730 that may be resident either on the wireless unit 720 itself, or separately from the wireless unit 720, as described with reference to FIG. 6. When residing separately from the wireless unit 720, the processing module 730 may be configured to run algorithms, perform computations, or perform anomaly checking that may be too complex or intensive for a low-power wireless unit 720 to implement. Such anomaly checking may correspond to the “self-checking” techniques as further described hereinbelow with reference to FIG. 8.

Processing module 730 includes a pre-processing block 735 that identifies and processes anomalies in the multi-channel signal. The output of pre-processing block 735 is provided to a data analysis block 737, which may output an event indicator 730a. The output event indicator 730a may correspond to the output of the estimation of the patient's brain state as described with reference to FIG. 6.

In the exemplary embodiment shown, the pre-processing block 735 communicates with an anomaly data service 740. The anomaly data service 740 may reside remotely from the processing module 730, and may provide the pre-processing block 735 with dynamically adjusted thresholds and/or other parameters to aid the pre-processing block 735 in identifying anomalies in the multi-channel signal. For example, the anomaly data service 740 may analyze anomalies from a plurality of multi-channel signals sampled over a population of seizure detection systems, seizure prediction systems, and/or seizure counter-prediction systems. The anomaly data service 740 may periodically derive preferred DCN comparison thresholds for use in the individual real-time analysis system 700. In an exemplary embodiment, the real-time analysis system 700 may also upload data samples to the anomaly data service 740 to aid the anomaly data service 740 in deriving preferred thresholds.

In an exemplary embodiment, the anomaly data service 740 may communicate with the processing module 730 wirelessly. Alternatively, the anomaly data service 740 may communicate with the processing module 730 over a wired connection. In yet another exemplary embodiment, the anomaly data service 740 may be omitted altogether, and the pre-processing block 735 may simply rely on pre-programmed threshold values. Such exemplary embodiments are contemplated to be within the scope of the present disclosure.

FIG. 8 depicts an exemplary method 800 according to the present disclosure that may be implemented, e.g., using the system 700. Note the exemplary method is shown for illustrative purposes only, and is not meant to limit the scope of the present disclosure to any particular method disclosed.

At step 810, the wireless unit 720 and processing module 730 are powered on.

At step 820, the wireless unit 720 receives the multi-channel signal from, e.g., a plurality of biosensors such as implanted devices 62 in FIG. 6.

At step 830, anomaly pre-processor 735 in processing module 730 identifies the presence of boot-up anomalies in the multi-channel signal received at step 820. This step may also be termed “self-checking,” or “self-test diagnostics.”

In an exemplary embodiment, “boot-up” anomalies may be any anomalies identified in the multi-channel signal during an initial boot-up phase. The boot-up phase may correspond to a time when software in the processing module 730 is initialized, and/or other parameters of the system 700 are initially configured. For example, the boot-up phase may last for a fixed amount of time after the wireless unit 720 and processing module 730 are powered on at step 800.

In an exemplary embodiment, the identification of anomalies in the multi-channel signal may be performed using the DCN computation techniques earlier described herein with reference to FIG. 3A. However, anomaly identification need not be performed using DCN computation. Anomaly identification in the method 800 may generally be performed using any suitable anomaly detection metric or metrics derivable by one of ordinary skill in the art in light of the present disclosure. Such alternative exemplary embodiments are contemplated to be within the scope of the present disclosure.

At step 840, an anomaly central data service may be continuously updated during operation of the method 800 with appropriate thresholds and/or algorithms for detecting the presence of anomalies in the multi-channel signal. In an exemplary embodiment, the anomaly data service may update a series of thresholds T1, T2, etc., against which the data condition number (DCN) is compared to detect the presence of anomalies in the multi-channel signal. The anomaly data service may vary the value of such thresholds over time, based on, e.g., offline analysis of anomalies and associated anomaly metrics as computed over an entire population of multi-channel signals.

At step 850, operation of the system 700 proceeds with the processing module 730 processing the multi-channel signal, taking into account the information in the anomaly data service.

At step 860, the anomaly processor 735 checks for anomalies in the multi-channel signal during normal operation of the system 700. The checking at step 860 may be termed “adaptive” anomaly identification and processing, as contrasted with the “boot-up” anomaly identification and processing described with reference to step 830. Information about anomalies identified during step 860 may be used to update the anomaly data service, as illustrated by the return arrow from step 860 to step 840, and as earlier described with reference to block 740 hereinabove. The steps 840, 850, 860 may be continuously repeated during normal operation of the system 700. An advantage of the adaptive anomaly identification and processing techniques described herein is that they may be varied over an extended temporal context used to monitor the multi-channel signal, as compared to the one-time self-checking diagnostics provided during a boot-up phase.

In an exemplary embodiment, entries from the anomaly data service may also be removed from the data service if anomaly processor 735 determines that such anomalies are no longer applicable. Such exemplary embodiments are contemplated to be within the scope of the present disclosure.

FIG. 9A illustrates an exemplary plot of data condition number (DCN) versus time, corresponding to an operating scenario wherein two of the sensors used to generate a multi-channel signal are gradually electrically shorted together during the interval starting from Time=5 minutes to Time=10 minutes. The scenario shown may arise, e.g., when two adjacent connectors at a distal end of a lead assembly implanted in a patient are gradually shorted together, effectively forming a path for interconnect ionic migration.

As seen in FIG. 9A, the DCN increases to extremely high levels in response to the gradual electrical shorting, and takes on values as high as 10¹⁵ when two of the leads are eventually completely shorted together, and converging to their common average, after Time=10 minutes. This expected behavior of the DCN in response to a physical condition such as electrical shorting allows the DCN to be used as an effective detection metric for indicating when such conditions might arise in an actual scenario. For example, by detecting when the DCN exceeds a suitably high threshold, the error condition wherein there is electrode shortage may be flagged to an automated software module, or to a manual user. One of ordinary skill in the art will appreciate that alternative techniques may also be employed to detect such an error condition, e.g., pattern matching the observed DCN to expected behavior of the DCN in the presence of the error condition, such expected behavior being obtained through, e.g., prior simulation or historical data. Furthermore, through further processing, e.g., by computing and evaluating the DCN over specific subsets of channels of the multi-channel signal, the specific leads causing the electrical shortage problem may be specifically identified and dealt with.

FIG. 9B illustrates an exemplary plot of data condition number (DCN) versus time, corresponding to an operating scenario wherein there is a gradual loss of contact of a reference electrode used to sample the multi-channel signal, over the span of about three days. As seen in FIG. 9B, there is a noticeable upward trend in the DCN over the time interval shown. In an exemplary embodiment, the presence of such an upward trend in the DCN may be used to detect the condition corresponding to gradual loss of contact of an electrode. Such a condition may then be flagged to an automated software module, or to a manual user. One of ordinary skill in the art will appreciate that various techniques may be employed to quantify any upward trend in the DCN. For example, a low-pass filter may be applied to the DCN time-series, and absolute increases in the DCN over a suitably long time period may be assessed and compared to some threshold. Alternatively, other pattern matching techniques may also be employed. Such alternative exemplary embodiments are contemplated to be within the scope of the present disclosure.

FIG. 10A illustrates an exemplary embodiment of a graphical display interface 1000 for simultaneously displaying various types of information associated with the techniques of the present disclosure. Note the exemplary embodiment of FIG. 10A is shown for illustrative purposes only, and is not meant to limit the scope of the present disclosure to any particular graphical display interface shown. One of ordinary skill in the art will appreciate that the graphical display interface 1000 may be a “screen shot,” i.e., an instantaneous sample of the image on a computer screen displayed when a user (e.g., a technician) is using the disclosed graphical interface to analyze data. The graphical display may be generated by, e.g., computer software running on a general-purpose computer, and may be readily designed by one of ordinary skill in the art in light of the further description given hereinbelow. For example, in an exemplary embodiment, such computer software may be conveniently accessed via a Web browser.

In FIG. 10A, graphical display interface 1000 includes a graphical representation of a catalog configuration block 1010, a graphical representation of a legend 1020, and a series of time-series displays 1040. The time-series displays 1040 may include time series 1041, 1042, 1043, each associated with a corresponding subject identification number as denoted in the subject ID column 1030. In the exemplary embodiment shown, there are three subject ID's 0001, 0002, and 0003. One of ordinary skill in the art will appreciate that the techniques disclosed herein may readily be applied to display information associated with any number of subjects, per the requirements of the application.

The catalog configuration block 1010 includes a series of drop-down menus 1011, 1012, and 1013.

Drop-down menu 1012 allows the user to select a scheme for backdrop patterns used to highlight certain information contained in the time-series displays 1040. In an exemplary embodiment, the mapping between a type of information and a corresponding backdrop pattern used to highlight such information type may be as shown according to the legend 1020. The legend 1020 indicates backdrop patterns used to show an ‘interictal’ segment 1021, a ‘preseizure’ segment 1022 (e.g., a fixed preictal time interval), other segment 1023, dropout 1024 (e.g., lapsed or invalid data), and seizure event 1025. As shown in the time-series 1041, 1042, and 1043, the backdrop patterns identified in legend 1020 may be used to annotate the time-series displays 1040 with information in a concise and easily presented form for convenient analysis by a user of the graphical display interface 1000.

One of ordinary skill in the art will appreciate that various modifications to the backdrop pattern scheme shown are readily derivable in light of the present disclosure. For example, in a color graphical display interface (not shown), backdrop colors may be used in place of, or in addition to, the backdrop patterns shown. A color scheme may assign a distinct color to each of the following types of segments: “interictal” segments, “preseizure” segments, “other” segments, “dropouts” (e.g., lapsed or invalid data), and “seizure” events. Alternative exemplary embodiments may also include other types of information not explicitly shown in FIG. 10A. Such alternative exemplary embodiments are contemplated to be within the scope of the present disclosure.

Drop-down menu 1013 allows the user to select the quantity to be displayed on the vertical axes of the time-series displays 1040. For example, one selectable quantity may be a data condition number (DCN) as calculated for each subject according to the present disclosure, plotted versus time. Alternative types of quantities include, e.g., the channel-sum of line-lengths (sum of absolute deviations in time) or its spatiotemporal and normalized variants, which would be appropriate for large-scale temporal localization of seizures rather than EEG anomalies.

As further shown in FIG. 10A, time series 1041 is accompanied by a time event index 1041.1. In the exemplary embodiment shown, the time event index 1041.1 is shown in bold and underlined, and has a numerical value of 20108. The time event index 1041.1 may indicate a time index in the time series 1041 wherein the corresponding DCN value exceeds some pre-specified threshold. See, e.g., the portion of time series 1041 indicated by 1041.1a. As more clearly shown in FIG. 10B, by rolling a cursor 1080A over the time event index 1041.1, an auxiliary window 1090A may “pop up” in the same display. Alternatively, the time event index 1041.1 may be “hyperlinked,” such that placing the cursor 1080A over the time event index 1041.1 and clicking the index 1041.1 causes the auxiliary window 1090A to simultaneously pop up.

The information content, or type of event, shown in the auxiliary window 1090A may be selected via the drop-down menu 1011. In the exemplary embodiment shown in FIG. 10B, the auxiliary window 1090A may show, e.g., the portion of the original multi-channel signal (such as EEG data) from which the DCN value corresponding to the time event index 1041.1 was derived. The auxiliary window 1090A advantageously allows the graphical interface user to visually inspect the portion of the multi-channel signal giving rise to the above-threshold DCN value alongside the DCN time-series, and manually determine whether the identified portion is in fact anomalous.

Other pop-up events to display may include context menus that direct the user to the raw EEG data where an anomaly can be further inspected in full detail.

As shown for indices 1043.1 and 1043.2 associated with time-series 1043, there may be multiple points in a time series wherein a DCN value exceeds a pre-specified threshold, and thus there may generally be multiple event indices associated with each time-series.

Based on the teachings described herein, it should be apparent that an aspect disclosed herein may be implemented independently of any other aspects and that two or more of these aspects may be combined in various ways. In one or more exemplary embodiments, the functions described may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software, the functions may be stored on or transmitted over as one or more instructions or code on a computer-readable medium. Computer-readable media includes both computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A storage media may be any available media that can be accessed by a computer. By way of example, and not limitation, such computer-readable media can comprise RAM, ROM, EEPROM, CD/DVD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, solid-state flash cards or drives, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer. Also, any connection is properly termed a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of medium. Disk and disc, as used herein, includes compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk and Blu-Ray disc where disks usually reproduce data magnetically, while discs reproduce data optically with lasers. Combinations of the above should also be included within the scope of computer-readable media.

In this specification and in the claims, it will be understood that when an element is referred to as being “connected to” or “coupled to” another element, it can be directly connected or coupled to the other element or intervening elements may be present. In contrast, when an element is referred to as being “directly connected to” or “directly coupled to” another element, there are no intervening elements present.

A number of aspects and examples have been described. However, various modifications to these examples are possible, and the principles presented herein may be applied to other aspects as well. These and other aspects are within the scope of the following claims.

Claims

1. A method for displaying information to a user, the method comprising: accepting input from the user selecting a metric to be displayed, said selected metric being associated with a multi-channel signal derived from outputs of a plurality of sensors positioned to acquire physiological signals, wherein each channel of the multi-channel signal corresponds to a different one of the plurality of sensors;displaying with a processor and a computer graphical display interface at least one time-series plot of the selected metric associated with the multi-channel signal;using a backdrop pattern, indicating on the at least one time-series plot portions of the plot having at least one identified characteristic;for each of the at least one time-series plot displayed, displaying with the processor and the computer graphical display interface a time event index wherein the corresponding metric meets a predetermined condition, wherein the predetermined condition comprises the selected metric being greater than a predetermined threshold;accepting input from the user as to whether to display further information associated with the time event index; anddisplaying the further information associated with the time event index when the user so specifies.

2. The method of claim 1, the metric comprising a data condition number calculated for the multi-channel signal.

3. The method of claim 1, the at least one time-series plot comprising a plurality of time-series plots, each time-series plot associated with a subject ID.

4. The method of claim 1, the at least one identified characteristic including a portion of the time-series corresponding to a seizure segment.

5. The method of claim 1, the backdrop pattern comprising a color scheme, the color scheme representing “interictal” segments, “preseizure” segments, “other” segments, “dropouts,” and “seizure” events.

6. The method of claim 1, the input from the user as to whether to display further information associated with the time event index comprising positioning of a cursor over the time event index.

7. The method of claim 1, further comprising: accepting input from the user indicating a color scheme for the backdrop pattern to be used.

8. The method of claim 1, the displaying the further information associated with the time event index comprising: displaying a pop-up plot of a portion of the multi-channel signal corresponding to the time event index.

9. The method of claim 8, the multi-channel signal comprising a multi-channel EEG signal.

10. A method for detecting anomalies in a multi-channel signal, the method comprising: sampling the multi-channel signal over a time window;computing with a processor an anomaly metric for the multi-channel signal over the time window; andidentifying the presence of an anomaly based on the magnitude of the anomaly metric; the computing an anomaly metric comprising: computing a condition number of the multi-channel signal over the time window; andadjusting the condition number based on a parameter of the multi-channel signal to generate a data condition number (DCN); the multi-channel signal comprising a signal sampled from a plurality of electrodes positioned to acquire physiological signals from a body, the method further comprising:generating a DCN time series corresponding to a plurality of time windows;generating an anomaly log based on the DCN time series;merging anomalies in the anomaly log separated by less than a minimum separation to generate a modified anomaly log;identifying segments of the multi-channel signal corresponding to anomalies in the modified anomaly log; andoutputting time-expanded versions of the identified segments to a record; the identifying the presence of an anomaly comprising matching the DCN time series to at least one known pattern of DCN time series corresponding to an anomalous condition.

11. A method for detecting anomalies in a multi-channel signal, the method comprising: sampling the multi-channel signal over a time window;computing with a processor an anomaly metric for the multi-channel signal over the time window; andidentifying the presence of an anomaly by comparing the magnitude of the anomaly metric to an optimum threshold; the computing an anomaly metric comprising:computing a condition number of the multi-channel signal over the time window; andadjusting the condition number based on a parameter of the multi-channel signal to generate a data condition number (DCN); the method further comprising:generating a DCN time series corresponding to a plurality of time windows;generating an event log based on the DCN time series, the generating an event log comprising identifying an event as a contiguous set of DCN values greater than a candidate threshold;merging events in the event log separated by less than a minimum separation to generate a modified event log;repeating the steps of generating a DCN time series, generating an event log, and merging events using a plurality of candidate thresholds;generating a plot of number of events identified in a modified event log versus candidate threshold used;attempting to identify at least one inflection point in the generated plot; andsetting the optimum threshold to be the candidate threshold corresponding to the inflection point with the largest abscissa in the plot.

12. The method of claim 11, further comprising, when the attempt to identify at least one inflection point is unsuccessful: identifying a maximum number of events corresponding to a candidate threshold in the histogram; andsetting the optimum threshold to be a candidate threshold whose corresponding number of events is less than or equal to a fixed percentage of the maximum number of events.