`|||||||||||||||||||||||||||||||||U|||| |||||||||||||||||||||||||||||||||||||||||||| ||||||||
`
`S 200301m4428A1
`
`(19) Ulllted States
`(12) Patent Application Publication (10) Pub. N0.: US 2003/0004428 A1
`
` Plcss et al. (43) Pub. Date: Jan. 2, 2003
`
`
`(54) SEIZURE SENSING AND I)ETICC'I‘I()N
`USING AN IMPlAN'I‘ABLE DEVICE
`
`Publication Classification
`
`(76)
`
`Inventors: Benjamin D. Pless, Atherton, CA (US);
`Stephen rl‘. Archer. Sunnyvale, CA
`(US); Craig Baysinger, Livermore, CA
`(US); Barbara Gibb, Paio Alto, CA
`(US); Suresh K. Gurunmhnn, Santa
`Clara. CA (US); Bruce Kirkpatrick,
`Santa Clara, CA (US); Thomas K.
`Tclteng, Pleasant Hill, CA (US)
`
`Correspondence Address:
`NEUROPACE, INC.
`255 SANTA ANA COURT
`SUNNYVALE, CA 94085 (US)
`
`(21) Appl. No:
`
`”(896,092
`
`(22)
`
`Filed:
`
`Jun. 28, 2001
`
`1m. (:1? ....................................................... AGIN ms
`(51)
`(52) US. Cl.
`.............................................................. 600544
`
`(57)
`
`ABSTRACT
`
`A system and method for detecting and predicting neuro-
`logical events with an impla ntable device uses a relatively
`low—power central processing unit in connection with signal
`processing circuitry to identify features (including half
`waves) and calculate window-based characteristiCs (includ-
`ing line lengths and areas under the curve of the waveform)
`in an electrographic signal received from a patient’s brain.
`The features and window-based characteristics are combin-
`
`able in various ways according to the invention to detect and
`predict neurological events in real time, enabling responsive
`action by the implantable device.
`
`APPLE 1016
`
` 1
`
`APPLE 1016
`
`1
`
`

`

`Patent Application Publication
`
`Jan. 2, 2003 Sheet 1 0f 22
`
`US 2003/0004428 A1
`
`
`
`2
`
`

`

`Patent Application Publication
`
`Jan. 2, 2003 Sheet 2 0f 22
`
`US 2003/0004428 A1
`
`210
`
`
`
`3
`
`

`

`Patent Application Publication
`
`Jan. 2, 2003 Sheet 3 0f 22
`
`US 2003/0004428 A1
`
` Implanted
`
`Device
`
`Fig. 3
`
`4
`
`

`

`Patent Application Publication
`
`Jan. 2, 2003 Sheet 4 0f 22
`
`US 2003/0004428 A1
`
`410
`
`Control Module
`
`Detection
`
`Subsystem
`
`
`‘
`
`
`Memoryr
`
`Subsystem
`
`432/
`
`Power Supply
`
`Communication
`
`"\
`Subsystem
`
`
`Stimulation
`
`
`
`Subsystem
`
`'K 430
`
`
`
`
`
`
`
`Clock Supply
`
`
`
`If?
`
`Fig. 4
`
`5
`
`

`

`Patent Application Publication
`
`Jan. 2, 2003 Sheet 5 0f 22
`
`US 2003/0004428 A1
`
`422
`
`Detection Subsystem
`
`Data
`
`Interface
`
`Waveform
`
`Analyzer
`
`Sensing
`
`Front
`
`Fig. 5
`
`6
`
`

`

`patent Application publication
`
`Jan. 2,2003 Sheetfi of 22
`
`US 2003/0004428A1
`
`512
`
`Sensing Front End
`
`-< Digital mm
`
`Analogto
`
`-'II
`
`Converter Ir: Multiplexer
`
`ll
`
`-II
`
`-ul
`
`I
`
`II
`
`Amplifier
`
`7
`
`

`

`Patent Application Publication
`
`Jan. 2, 2003 Sheet 7 0f 22
`
`US 2003/0004428 A1
`
`Waveform Analyzer
`
`Wave
`
`Morphology
`
`Channel
`
`Event Detector
`
`Controlier
`
`8
`
`

`

`Jan. 2, 2003 Sheet 8 0f 22
`Patent Application Publication
`
`
`US 2003/0004428 A1
`
`31 2
`
`\\ Event Detector
`
`824.J
`
`
`
`814
`
`Channel 2
`
`834
`
`856
`/
`
`
`
`820
`
`
`
`9
`
`

`

`Patent Application Publication
`
`Jan. 2, 2003 Sheet 9 0f 22
`
`US 2003/0004428 A1
`
` f
`
`,3;
`I[:::::‘
`r'""k-|——————‘i
`
`
`10
`
`10
`
`

`

`Patent Application Publication
`
`Jan. 2, 2003 Sheet 10 0f 22
`
`US 2003/0004428 A1
`
`1040
`\
`
`11144
`t
`
`1048
`\
`
`\
`
`\
`
`1020
`x
`
`
`
`1052
`
`\
`
`1038
`1050
`1042 \ 1046
`\1.
`\ \ \ .\‘\\~: \
`10.13
`\' ’
`
`
`
`{211-1032 1035
`|
`
`
`1030
`
`Fig. 10
`
`11
`
`11
`
`

`

`Patent Application Publication
`
`Jan. 2, 2003 Sheet 11 0f 22
`
`US 2003/0004428 A1
`
`Begin
`
`Initialize WaveTime
`
`/
`
`11,10
`
`,
`
`Fig. 11
`
`Initialize WaveDur,
`EndThreshold, Peak,
`
`FirstSample
`
`Wait for
`
`.
`
`EEG Sample
`
`Increment
`
`WaveTime,
`WaveDur
`
`EndThreshotd,
`Peak
`
`/ 1132
`
`Initialize WaveDur,
`Endlhreshold, Peak,
`FirstSample
`
`Increment
`
`WaveTime,
`WaveDur
`
`
`
`
`
` Reset
`
`EndThreshold,
`Peak
`
`
`
`
`
`
`
`
`
` Reset
`
`Calculate
`ll
`Aguprgtlilgr?
`
`\
`\
`\ 1124
`
`Calculate
`Amplitude,
`Duration
`
`5/1126
`
`
`Amp and Dur
`Quallfy?
`
`E23
`
`Store
`
`
`
`11318
`/
`
`Amp and Dur
`'
`
`~
`
`
`
`
`Reset
`WaveTime
`
`
`12
`
`12
`
`

`

`Patent Application Publication
`
`Jan. 2, 2003 Sheet 12 0f 22
`
`US 2003/0004428 A1
`
`Fig 1 2
`
`/1210
`
`/1212
`
`Clear
`
`Haltwave
`
`Window Flag
`
`Identity New
`Qualified
`Halfwaves
`
`Set Current
`
`Halfwave to Oldest
`
`New Halfwave
`
`1216
`
`Test Interval Between
`
`Current Halfwave and
`
`Current Halfwave - x
`
`Interval < H.W. '
`
`Window?
`
`Increment
`
`Current
`Halfwave
`
`/1220
`
`Set
`Haltwave
`
`Window Flag
`
`,1214
`
`
`
`1222
`
`1 224
`
`Apply
`Logic
`Inversion
`
`More New
`Qualified
`Halfwaves?
`
`
`
`13
`
`13
`
`

`

`Patent Application Publication
`
`Jan. 2, 2003 Sheet 13 0f 22
`
`US 2003/0004428 A1
`
`Fig. 13
`
`1318\
`
`Decrement
`
`
`
`/1310
`
`Clear Sum;
`Initialize
`
`Window Pointer
`
`Window
`
`Pointer
`
`Wave Window
`
`Flag Set?
`
`
`
`
`
`Increment
`Sum
`
` More
`Windows?
`
`1 322
`
`
`
`Apply
`Persistence
`
`/1326
`
`14
`
`14
`
`

`

`Patent Application Publication
`
`Jan. 2, 2003 Sheet 14 0f 22
`
`US 2003/0004428 A1
`
`
`
`15
`
`15
`
`

`

`Patent Application Publication
`
`Jan. 2, 2003 Sheet 15 0f 22
`
`US 2003/0004428 A1
`
`Fig. 15
`
`Initialize Total
`
`Store Total
`
`Wait for
`
`Current
`
`Sample
`
`Difference =
`
`| Current Sample -
`Previous Sample |
`
`Previous Sample =
`Current Sample
`
`Add Difference to
`
`Total
`
`Last Sample
`in Window?
`
`16
`
`16
`
`

`

`Patent Application Publication
`
`Jan. 2, 2003 Sheet 16 0f 22
`
`US 2003/0004428 A1
`
`1610 \
`
`
`
`Initialize
`Accumulated
`
`Fig. 16
`
`1612
`
`L.L. Total
`
`
`\ Current Window =
`Calculate
`Most Recent
`LL. Trend
`Window - n + 1
`
`1622
`
`/
`
`
`
`
` Clear Line
`
`Length Flag
`
`L.L.
`
`Trend Interval
`
`
`Passed?
`
`'
`
`x 1614
`
`Add Line Length of 1
`Current Window to
`Accumulated LL.
`Total
`
`Calculate
`LL. Threshold
`
`Increment
`current
`Window
`
`L.L. Total Exceed
`Threshold?
`
`X 1628
`
`Length Flag
`
`A I
`Lipi:
`g -
`lnverston
`
`(/1634
`
`Apply
`Persistence
`
`17
`
`17
`
`

`

`Patent Application Publication
`
`Jan. 2, 2003 Sheet 17 0f 22
`
`US 2003/0004428 A1
`
`2§.W/zfiyr/V/WM
`
`§§§
`
`171?
`
`§/§
`
`Fig. 17
`
`18
`
`18
`
`
`
`

`

`Patent Application Publication
`
`Jan. 2, 2003 Sheet 18 of 22
`
`US 2003/0004428 A1
`
`Initialize Total
`
`Area =
`
`| Current Sample |
`
`Store Total
`
`Add Area to Total
`
`Last Sample
`in Window?
`
`Fig. 18
`
`19
`
`19
`
`

`

`Total
`
`
`Current Window =
`Most Recent
`
`Add Area of
`Current Window to I
`Accumulated Area
`
`Increment
`Current
`Window
`
` Area
`
`Trend Interval
`Passed?
`
`
`Y
`
`Patent Application Publication
`
`Jan. 2, 2003 Sheet 19 0f 22
`
`US 2003/0004428 A1
`
`Initialize
`Accumulated
`Area Total
`
`Window- n +‘I
`
`
`
`Calculate
`Area Trend
`
` Calculate
`Area Threshold
`
`
`
`
`f/
`
`Totai Exceed Area
`Threshold?
`
`1 928
`
`Set Area Flag
`
`Ciear
`
`Area Fiag
`
`
`
`Apply
`Logic
`Inversion
`
`APP'Y
`Persistence
`
`20
`
`20
`
`

`

`Patent Application Publication
`
`Jan. 2, 2003 Sheet 20 0f 22
`
`US 2003/0004428 A1
`
`2010
`
`2018
`
`Receive
`' Combine
`
`Interrupt
`Detection Channels
`Anaiysis Tools into
`
` 2020
`
`Combine
`
`Detection Channels
`into Event Detectors
`
`2012
`
`2014
`
`2024
`
`
`Check Line
`
`Lengths
`
`2016\
`
`Trigger
`
`Ftag Set?
`
` Check Areas
`
`
`initiate Action
`
`
`
`Fig. 20
`
`21
`
`21
`
`

`

`Patent Application Publication
`
`Jan. 2, 2003 Sheet 21 0f 22
`
`US 2003/0004428 A1
`
`[2110
`
`
`Set Output
`Flag
`
`
`2112
`
`Start with First
`
`Analysis Tool
`
`
`
`
`2116
`
`2120
`
`/
`
`Clear Output
`Flag
`
`Apply
`Logic
`inversion
`
`
`More
`Analysis
`Tools?
`
`Fig. 21
`
`22
`
`22
`
`

`

`Patent Application Publication
`
`Jan. 2, 2003 Sheet 22 0f 22
`
`US 2003/0004428 A1
`
`2210
`
`Set Output
`Flag
`
`Start with First
`
`Detection
`
`Channel
`
`
`
`
`2214
`
`Detection Channel
`
`Detection Channel
`
`Clear Output
`Flag
`
`
`
`
`
`
`
`More
`
`
`Detection Ch anneis
`?
`
`2220
`
`Fig . 22
`
`2222
`
`Increment
`
`Channel
`
`23
`
`23
`
`

`

`US 2003/0004428 A1
`
`Jan. 2, 2003
`
`SEIZURE SENSING ANI) DETECTION USING AN
`IMPLANTABLE DEVICE
`
`FIELD OF THE INVENTION
`
`[0001] The invention relates to systems and methods for
`detecting and predicting neurological dysfunction charac-
`terized by abnormal electrographic patterns, and more par-
`ticularly to a system and method for detecting and predicting
`epileptic seizures. and their onsets by analyzing electroen-
`cephalogram and
`electrocortioogram signals with an
`implantable device.
`
`BACKGROUND OF THE INVENTION
`
`[0002] Epilepsy, a neurological disorder characterized by
`the occurrence of seizures (specifically episodic impairment
`or loss of consciousness, abnormal motor phenomena, psy-
`chic or sensory disturbances, or the perturbation of the
`autonomic nervous system), is debilitating to a great number
`of people. It is believed that as many as two to four million
`Americans may su tfer
`from various forms of epilepsy.
`Research has found that its prevalence may be even greater
`worldwide, particularly in less economically deveIOped
`nations, suggesting that the worldwide figure for epilepsy
`sufferers may be in excess of one hundred million.
`
`[0003] Because epilepsy is characterized by seizures, its
`sufferers are frequently limited in the kinds of activities they
`may participate in. Epilepsy can prevent people from drivw
`ing, working, or otherwise participating in much of what
`society has to olfer. Some epilepsy sulIerers have serious
`seizures so frequently that they are effectively incapacitated.
`
`Furthermore, epilepsy is often progressive and can
`[0004]
`be associated with degenerative disorders and conditions.
`Over time, epileptic seizures often become more frequent
`and more serious, and in particularly severe cases, are likely
`to lead to deterioration of other brain functions (including
`cognitive function) as well as physical impairments.
`
`[0005] The current state of the art in treating neurological
`disorders, particularly epilepsy,
`typically involves drug
`therapy and surgery. The first approach is usually drug
`therapy.
`
`[0006] A number of drugs are approved and available for
`treating epilepsy, such as sodium valproate, phenobarbitabr
`primidone, ethosuximide, gabapentin, phenytoin, and car-
`bamazepine, as well as a number of others. Unfortunately,
`those drugs typically have serious side effects, especially
`toxicity, and it
`is extremely important
`in most cases to
`maintain a precise therapeutic serum level to avoid break
`through seizures (if the dosage is too low) or toxic effects {if
`the dosage is too high). The need for patient discipline is
`high, especially when a patient’s drug regimen causes
`unpleasant side effects the patient may wish to avoid.
`
`[0007] Moreover, while many patients respond well to
`drug therapy alone, a significant number (at least 20-30%)
`do not. For those patients, surgery is presently the best~
`established and most viable alternative course of treatment.
`
`[0008] Currently practiced surgical approaches include
`radical surgical resection such as hemisphereetomy, corti-
`cectomy, lohectomy and partial lobectomy, and less-radical
`lesionectomy, transection, and stereotactic ablation. Besides
`being less than fully successful, these surgical approaches
`
`generally have a high risk of complications, and can often
`result in damage to eloquent {i.e., functionally important)
`brain regions and the consequent long—term impairment of
`various cognitive and other neurological functions. Further—
`more, for a variety of reasons, such surgical treatments are
`contraindicated in a substantial number of patients. And
`unfortunately, even after radical brain surgery, many epi-
`lepsy patients are still not seizure-free.
`
`[0009] Electrical stimulation is an emerging therapy for
`treating epilepsy. However, currently approved and avail—
`able electrical stimulation devices apply continuous electri-
`cal
`stimulation to neural
`tissue surrounding or near
`implanted electrodes, and do not perform any detection—
`they are not responsive to relevant neurological conditions.
`
`from
`(NCP)
`[0010] The NeuroCybernetic Prosthesis
`Cyberonics,
`for example, applies continuous electrical
`stimulation to the patient’s vagus nerve. This approach has
`been found to reduce seizures by about 50% in about 50%
`of patients. Unfortunately, a much greater reduction in the
`incidence of seizures is needed to provide clinical benefit.
`The Activa device from Medtronic is a pectorally implanted
`continuous deep brain stimulator intended primarily to treat
`Parkinson’s disease. In operation, it supplies a continuous
`electrical pulse stream to a selected deep brain structure
`where an electrode has been implanted.
`
`[0011] Continuous stimulation ofdeep brain structures for
`the treatment of epilepsy has not met with consistent suc-
`cess. To be effective in terminating seizures, it is believed
`that one effective site where stimulation should be per-
`formed is near the focus of the epiteptogenic region. The
`focus is often in the neomrtex, where continuous stimula-
`tion may cause significant neurological deficit with clinical
`symptoms including loss of speech, sensory disorders, or
`involuntary motion. Accordingly, research has been directed
`toward automatic responsive epilepsy treatment based on a
`detection of imminent seizure.
`
`[0012] A typical epilepsy patient experiences episodic
`attacks or seizures, which are generally electrographically
`defined as periods of abnormal neurological activity. As is
`traditional in the art, such periods shall be referred to herein
`as “total”.
`
`[0013] Most prior work on the detection and responsive
`treatment of seizures via electrical stimulation has foeused
`on analysis of electroencephalogram (EEG) and electrocor—
`ticogram (ECoG) waveforms. In general, EEG signals rep—
`resent aggregate neuronal activity potentials detectable via
`electrodes applied to a patient's scalp. ECoG signals, deep
`brain counterparts to EEG signals, are detectable via elec—
`trodes implanted on or under the dura mater, and usually
`within the patient’s brain. Unless the context clearly and
`expressly indicates otherwise, the term “EEG” shall be used
`generically herein to refer to both EEG and ECoG signals.
`
`[0014] Much of the work on detection has focused on the
`use of time-domain analysis of EEG signals. See, e.g., J.
`Gotman, Automatic seizure detection:
`improvements and
`evaluation, Electroencephalogr. Clin. Neurophysiol. 1990;
`'r'6(4): 317—24. In a typical time—domain detection system,
`EEG signals are received by one or more implanted elec-
`trodes and then processed by a control module, which then
`is capable of performing an action (intervention, warning,
`recording, etc.) when an abnormal event is detected.
`
`24
`
`24
`
`

`

`US 2003/0004428 A1
`
`Jan. 2, 2003
`
`is generally preferable to be able to detect and
`It
`[0015]
`treat a seizure at or near its beginning, or even before it
`begins. The beginning of a seizure is referred to herein as an
`"onset." However, it is important to note that there are two
`general varieties of seizure onsets. A“clinical onset" repre-
`sents the beginning of a seizure as manifested through
`observable clinical symptoms, such as involuntary muscle
`movements or neurophysiological elfects such as lack of
`responsiveness. An "electrographic onset" refers to the
`beginning of detectable electrographic activity indicative of
`a seizure. An electrographic onset will frequently occur
`before the corresponding clinical onset, enabling interven~
`tion before the patient suffers symptoms, but
`that
`is not
`always the case. In addition, there are changes in the EEG
`that occur seconds or even minutes before the electrographic
`onset that can be identified and used to facilitate intervention
`
`before electrographic or clinical onsets occur. This capabil-
`ity would be considered seizure prediction, in contrast to the
`detection of a seizure or its onset.
`
`In the Gotman system, EEG waveforms are filtered
`[0016]
`and decomposed into “features” representing characteristics
`of interest in the waveforms. One such feature is character-
`ized by the regular occurrence {i.e., density) of half-waves
`exceeding a threshold amplitude occurring in a specified
`frequency hand between approximately 3 Hz and 20 Hz,
`especially in comparison to background (non-ictal) activity.
`When such half~waves are detected,
`it
`is believed that
`seizure activity is occurring For related approaches, see also
`ll. Qu and J. Gotman, Aseizure warning system for long
`term epilepsy monitoring, Neurology 1995; 45: 2250-4; and
`ll. Qu and .l. (lotman, A Patient-Specific Algorithm for the
`Detection of Seizure Onset in Long-Term EEG Monitoring:
`Possible Use as a Warning Device, IEEE Trans. Biomed.
`Eng. 1997; 44(2): 115-22.
`
`[0017] The German articles address half wave character-
`istics in general, and introduce a variety of measurement
`criteria, including a ratio of current epoch amplitude to
`background; average current epoch EEG frequency; average
`background EEG frequency; coefficient of variation of wave
`duration; ratio of current epoch amplitude to following time
`period; average wave amplitude; average wave duration;
`dominant frequency (peak frequency of the dominant peak);
`and average power in a main energy zone. These criteria are
`variously mapped into an n-dimensional space, and whether
`a seizure is detected depends on the vector distance between
`the parameters of a measured segment of EEG and a seizure
`template in that space.
`
`It should be noted that the schemes set forth in the
`[0018]
`above articles are not
`tailored for use in an implantable
`device, and hence typically require more computational
`ability than would be available in such a device.
`
`[0019] U.S. Pat. No. 6,018,682 to Rise describes an
`implantable seizure warning system that implements a form
`of the Gotman system. However,
`the system described
`therein uses only a single detection modality, namely a count
`of sharp spike and wave patterns within a timer period. This
`is accomplished with relatively complex processing, includ-
`ing averaging over time and quantifying sharpness by way
`ofa second derivative of the signal. The Rise patent does not
`disclose how the signals are processed at a low level, nor
`does it explain detection criteria in any sufliciently specific
`level of detail.
`
`[0020] A more computationally demanding approach is to
`transform EEG signals into the frequency domain for rig-
`orous spectrum analysis. See, e.g., US. Pat. No. 5,995 ,868
`to Dorfineister et al., which analyzes the power spectral
`density of EEG signals in comparison to background char-
`acteristics. Although this approach is generally believed to
`achieve good results, for the most part, its computational
`expense renders it less than optimal for use in long-term
`implanted epilepsy monitor and treatment devices. With
`current technology, the battery life in an implantable device
`computationally capable of performing the Dorfmeister
`method would be too short for it to be feasible.
`
`[0021] Also representing an alternative and more complex
`approach is U.S. Pat. No. 5,857,978 to I'Iively et al., in which
`various non-linear and statistical characteristics of EEG
`signals are analyzed to identify the onset of ictal activity.
`Once more, the calculation of statistically relevant characA
`teristics is not believed to be feasible in an implantable
`device.
`
`[0022] US. Pat. No. 6,016,449 to Fischell, et al. (which is
`hereby incorporated by reference as though set forth in full
`herein), describes an implantable seizure detection and treat-
`ment system.
`In the Fischell system, various detection
`methods are possible, all of which essentially rely upon the
`analysis {either in the time domain or the frequency domain)
`ot‘processed EEG signals. Fischell ‘s controller is preferably
`implanted intracranially, but other approaches are also pos»
`sible, including the use of an external controller. When a
`seizure is detected, the Fischell system applies responsive
`electrical stimulation to terminate the seizure, a capability
`that will he discussed in further detail below.
`
`[0023] All ofthese approaches provide useful information,
`and in some cases may provide sufficient
`information for
`accurate detection and prediction of most imminent epileptic
`seizures.
`
`[0024] However, none of the various implementations of
`the known approaches provide 100% seizure detection accu-
`racy in a clinical environment.
`
`[0025] Two types of detection errors are generally pos-
`sible. A“false positive," as the term is used herein, refers to
`a detection of a seizure or ictal activity when no seizure or
`other abnormal event
`is actually occurring. Similarly, a
`"false negative“ herein refers to the failure to detect a seizure
`or ictal activity that actually is occurring or shortly will
`occur.
`
`In most cases, with all known implementations of
`[0026]
`the known approaches to detecting abnormal seizure activity
`solely by monitoring and analyzing EEG activity, when a
`seizure detection algorithm is tuned to catch all seizures,
`there will be a significant number of false positives. While
`it
`is currently believed that there are minimal or no side
`effects to limited amounts of over-stimulation (e .g., provid-
`ing stimulation sufficient to terminate a seizure in response
`to a false positive), the possibility of accidentally initiating
`a seizure or increasing the patient’s susceptibility to seizures
`must be considered.
`
`[002?] As is well known, it has been suggested that it is
`possible to treat and terminate seizures by applying electri-
`cal stimulation to the brain. See, e.g., US. Pat. No. 6,016,
`449 to Fischell et al., and II. R. Wagner, et al., Suppression
`ofcortical epileptiform activity by generalized and localized
`
`25
`
`25
`
`

`

`US 2003/0004428 A1
`
`Jan. 2, 2003
`
`L»)
`
`ECoG desynchronization, Electroencephalogr. Clin. Neuro—
`physiol. 1975; 39(5): 499-506. And as stated above,
`it is
`believed to be beneficial to perform this stimulation only
`when a seizure (or other undesired neurological event) is
`occurring or about
`to occur, as inappropriate stimulation
`may result in the initiation of seizures.
`
`Furthermore, it should be noted that a false nega-
`[0028]
`tive (that is, a seizure that occurs without any warning or
`treatment from the device) will often cause the patient
`significant discomfort and detriment. Clearly, false nega-
`tives are to be avoided.
`
`to achieve an
`It has been found to be difficult
`[0029]
`acceptably low level of false positives and false negatives
`with the level of computational ability available in an
`implantable device with reasonable battery life.
`
`the battery in an implantable device,
`Preferably,
`[0030]
`particularly one implanted intracranially, should last at least
`several years. There is a substantial risk of complications
`(such as infection, blood clots, and the overgrowth of scar
`tissue) and lead failure each time an implanted device or its
`battery is replaced. Rechargeable batteries have not been
`found to provide any advantage in this regard, as they are not
`as ellicient as traditional cells, and the additional electronic
`circuitry required to support the recharging operation con-
`tributes to the device ‘5 size and complexity. Moreover, there
`is a need for patient discipline in recharging the device
`batteries, which would require the frequent transmission of
`a substantial amount of power over a wireless link and
`through the patient’s skin and other tissue.
`[0031] As stated above,
`the detection and prediction of
`ictal activity has traditionally required a significant amount
`of computational ability. Moreover, for an implanted device
`to have significant real-world utility, it is also advantageous
`to include a number of other features and capabilities.
`Specifically,
`treatment (via electrical stimulation or drug
`infusion) andlor warning (via an audio annunciator,
`for
`example), recording of EEG signals for later consideration
`and analysis, and telemetry providing a link to external
`equipment are all useful capabilities for an implanted device
`capable of detecting or predicting epileptiform signals. All
`of these additional subsystems will consume further power.
`[0032] Moreover, size is also a consideration. For various
`reasons,
`intracranial
`implants
`are
`favored. A device
`implanted intracranially (or under the scalp) will typically
`have a lower risk of failure than a similar device implanted
`pectorally or elsewhere, which require a lead to be run from
`the device,
`through the patient ’s neck to the electrode
`implantation sites in the patient’s head. This lead is also
`prone to receive additional electromagnetic interference.
`[0033] As is well known in the art, the computational
`ability of a processor-controlled system is directly related to
`both size and power consumption. In accordance with the
`above considerations, therefore, it would be advantageous to
`have sulficient detection and prediction capabilities to avoid
`a substantial number of false positive and false negative
`detections, and yet consume little enough power (in con—
`junction with the other subsystems) to enable long battery
`life. Such an implantable device would have a relatively
`low—power central processing unit to reduce the electrical
`power consumed by that portion.
`[0034] At the current time, there is no known implantable
`device that is capable of detecting and predicting seizures
`
`and yet has adequate battery life and the consequent accept—
`ably low risk factors for use in human patients.
`
`SUMMARY OF THE INVENTION
`
`[0035] Accordingly, an implantable device according to
`the invention for detecting and predicting epileptic seizures
`includes a relatively low-speed and low-power central pro-
`cessing unit, as well as customized electronic circuit mod-
`ules in a detection subsystem. As described herein,
`the
`detection subsystem also performs prediction, which in the
`context of the present application is a form of detection that
`occu rs before identifiable clinical symptoms or even obvious
`electrographic patterns are evident upon inspection. The
`same methods, potentially with different parameters, are
`adapted to be used for both detection and prediction. Gen-
`erally, as described herein, an event (such as an epileptic
`seizure) may be detected, an electrographic "onset” of such
`an event (an electrographic indication of an event occurring
`at the same time as or before the clinical event begins) may
`be detected (and may be characterized by different wave
`form observations than the event itself), and a "precursor" to
`an event (electrographic activity regularly occurring some
`time before the clinical event) may be detected as predictive
`of the event.
`
`[0036] As described herein and as the terms are generally
`understood, the present approach is generally not statistical
`or stochastic in nature. The invention, and particularly the
`detection subsystem thereof, is specifically adapted to per—
`form much of the signal processing and analysis requisite for
`accurate and elfective event detection. The central process
`ing unit remains in a suspended “sleep" state characterized
`by relative inactivity a substantial percentage of the time,
`and is periodically awakened by interrupts from the detec—
`tion subsystem to perform certain tasks related to the detec-
`tion and prediction schemes enabled by the device.
`
`[0037] Much of the processing performed by an implant-
`able system according to the invention involves operations
`on digital data in the time domain. Preferably, to reduce the
`amount of data processing required by the invention,
`samples at ten—bit resolution are taken at a rate less than or
`equal to approximately 500 Hz (2 ms per sample).
`
`[0038] As stated above. an implantable system according
`to the invention is capable of accurate and reliable seizure
`detection and prediction. To accomplish this, the invention
`employs a combination of signal processing and analysis
`modalities, including data reduction and feature extraction
`techniques, mostly implemented as customized digital elec—
`tronics modules. minimally reliant upon a central processing
`unit.
`
`In particular, it has been found to be advantageous
`[0039]
`to utilize two different data reduction methodologies, both of
`which collect data representative of EEG signals within a
`sequence of uniform time windows each having a specified
`duration.
`
`[0040] The first data reduction methodology involves the
`calculation of a "line length function" for an EEG signal
`within a time window. Specifically, the line length function
`of a digital signal represean an accumulation of the sample-
`to-sample amplitude variation in the EEG signal within the
`time window. Stated another way, the line length function is
`representative of the variability of the input signal. A con-
`
`26
`
`26
`
`

`

`US 2003/0004428 A1
`
`Jan. 2, 2003
`
`stant input signal will have a line length of zero (represen—
`tative of substantially no variation in the signai amplitude),
`while an input signal that oscillates between extrema from
`sample to sample will approach the maximum line length. It
`should be noted that while the line length function has a
`physical—world analogue in measuring the vector distance
`traveled in a graph of the input signal, the concept of line
`length as treated herein disregards the horizontal (X) axis in
`such a situation. The horivontal axis herein is representative
`oftime, which is not combinablc in any meaningful way in
`accordance with the invention with information relating to
`the vertical (Y) axis, generally representative of amplitude,
`and which in any event would contribute nothing of interest.
`
`[0041] The second data reduction methodology involves
`the calculation of an “area function” represented by an EEG
`signal within a time window. Specifically, the area function
`is calculated as an aggregation of the EEG’s signal
`total
`deviation from zero over the time window, whether positive
`or negative. The mathematical analogue for the area function
`defined above is the mathematical integral of the absolute
`value of the EEG function (as both positive and negative
`signals contribute to positive area). Once again, the hori—
`zontal axis (time) makes no contribution to accumulated
`energy as treated herein. Accordingly, an input signal that
`remains around zero will have a small area value, while an
`input signal that remains around the most-positive or most-
`negative values will have a high area value.
`
`[0042] Both the area and line length functions may
`undergo linear or non—linear transformations. An example
`would be to square each amplitude before summing it in the
`area function. This non-linear operation would provide an
`output that would approximate the energy of the signal for
`the period of time it was integrated. Likewise linear and
`non-linear transformations of the difierence between sample
`values are advantageous in customizing the line length
`function to increase the effectiveness of the detector for a
`specific patient.
`
`[0043] The central processing unit receives the line length
`function and area function measurements performed by the
`detection subsystem, and is capable ofacting based on those
`measurements or their trends.
`
`Feature extraction, specifically the identification of
`[0044]
`half waves in an EEG signal, also provides useful informa-
`tion. A half wave is an interval between a local waveform
`minimum and a local waveform maximum; each time a
`signal "changes directions” (from increasing to decreasing,
`or vice versa), subject to limitations that will be set forth in
`further detail below, a new half wave is identified.
`
`[0045] The identification of half waves having specific
`amplitude and duration criteria allows some frequency-
`driven characteristics of the EEG signal to be considered and
`analyzed without the need for computationally intensive
`transformations of normally time-domain EEG signals into
`the frequency domain. Specifically, the half wave feature
`extraction capability of the invention identifies those half
`waves in the input signal having a duration that exceeds a
`minimum duration criterion and an amplitude that exceeds a
`minimum amplitude criterion. The number of half waves in
`a time window meeting those criteria is somewhat repre-
`sentative of the amount of energy in a waveform at a
`frequency below the frequency corresponding to the mini-
`mum duration criterion. And the number of half waves in a
`
`time window is constrained somewhat by the duration of
`each halfwave (i.e., if the half waves in a time window have
`particularly long durations, relatively fewer of them will fit
`into the time window),
`that number is highest when a
`dominant waveform frequency most closely matches the
`frequency corresponding to the minimum duration criterion.
`
`[0046] As stated above, the half waves, line length func-
`tion, and area fu nction ofvarious EEG signals are calculated
`by customized electronics modules with minimal involve-
`ment by the central processing unit, and are selectively
`combined by a system according to the invention to provide
`detection and prediction of seizure activity, so that appro-
`priate action can then be taken.
`
`[0047] Accordingly, in one embodiment of the invention,
`a system according to the invention includes a central
`processing unit, a detection subsystem located therein that
`includes a waveform analyzer. The waveform analyzer
`includes waveform feature analysis capabilities (such as half
`wave characteristics) as well as window-based analysis
`capabilities (such as line length and area under the curve),
`and both aspects are combined to provide enhanc